Framework for Content Distribution Network Interconnection (CDNI)Akamai Technologies, Inc.8 Cambridge CenterCambridgeMA02142USAlapeters@akamai.comVMware, Inc.3401 Hillview Ave.Palo AltoCA94304USAbdavie@vmware.comTNOBrassersplein 2Delft2612CTthe Netherlands+31-88-866-7000ray.vanbrandenburg@tno.nlThis document presents a framework for Content Distribution Network
Interconnection (CDNI). The purpose of the framework is to provide an
overall picture of the problem space of CDNI and to describe the
relationships among the various components necessary to interconnect
CDNs. CDNI requires the specification of interfaces and
mechanisms to address issues such as request routing, distribution
metadata exchange, and logging information exchange across CDNs. The
intent of this document is to outline what each interface needs to
accomplish and to describe how these interfaces and mechanisms fit
together, while leaving their detailed specification to other documents.
This document, in combination with RFC 6707, obsoletes RFC 3466.This document provides an overview of the various components
necessary to interconnect CDNs, expanding on the problem statement and
use cases introduced in and . It describes the necessary interfaces and mechanisms
in general terms and outlines how they fit together to form a complete
system for CDN Interconnection. Detailed specifications are left to
other documents. This document makes extensive use of message flow
examples to illustrate the operation of interconnected CDNs, but these
examples should be considered illustrative rather than prescriptive. uses different terminology and models for
"Content (distribution) Internetworking (CDI)". It is also less prescriptive in terms
of interfaces. To avoid confusion, this document obsoletes .This document uses the core terminology defined in . It also introduces the following terms: a hostname (Fully Qualified Domain Name -- FQDN) at the beginning of a URL (excluding
port and scheme), representing a set of content that is served by a
given CDN. For example, in the URL http://cdn.csp.example/...rest of
URL..., the CDN-Domain is cdn.csp.example. A major role of CDN-Domain
is to identify a region (subset) of the URI space relative to which
various CDNI rules and policies apply. For
example, a record of CDNI Metadata might be defined for the set of
resources corresponding to some CDN-Domain. a CDN-Domain that is allocated by a CDN
for the purposes of communication with a peer CDN but that is not
found in client requests. Such CDN-Domains may be used for inter-CDN
acquisition, or as redirection targets, and enable a CDN to
distinguish a request from a peer CDN from an end-user request. the CDN that ultimately delivers a piece of content
to the end user. The last in a potential sequence of Downstream
CDNs. When an Upstream CDN elects to
redirect a request towards a Downstream CDN, the Upstream CDN can base
its redirection purely on a local decision (and without attempting to
take into account how the Downstream CDN may in turn redirect the user
agent). In that case, the Upstream CDN redirects the request to the
Request Routing system in the Downstream CDN, which in turn will
decide how to redirect that request: this approach is referred to as
"Iterative" CDNI Request Redirection. When an Upstream CDN elects to
redirect a request towards a Downstream CDN, the Upstream CDN can
query the Downstream CDN Request Routing system via the CDNI Request
Routing Redirection interface (or use information cached from earlier
similar queries) to find out how the Downstream CDN wants the request
to be redirected. This allows the Upstream CDN to factor in the
Downstream CDN response when redirecting the user agent. This approach
is referred to as "Recursive" CDNI Request Redirection. Note that the
Downstream CDN may elect to have the request redirected directly to a
Surrogate inside the Downstream CDN, or to any other element in the
Downstream CDN (or in another CDN), to handle the redirected request
appropriately. operations between CDNs that happen
during the process of servicing a user request, i.e., between the time
that the user agent begins its attempt to obtain content and the time
at which that request is served. operations between CDNs that happen
independently of any given user request, such as advertisement of
footprint information or pre-positioning of content for later
delivery. a subset of the CDNI Control interface that
includes operations to pre-position, revalidate, and purge both
metadata and content. These operations are typically called in
response to some action (Trigger) by the Content Service Provider
(CSP) on the Upstream CDN.We also sometimes use uCDN and dCDN as shorthand for Upstream CDN
and Downstream CDN (see ), respectively.At various points in this document, the concept of a CDN footprint
is used. For a discussion on what constitutes a CDN footprint, the
reader is referred to .This document uses the reference model in ,
which expands the reference model originally defined in . (The difference is that the expanded model splits
the Request Routing interface into its two distinct parts: the Request
Routing Redirection interface and the Footprint & Capabilities
Advertisement interface, as described below.)While some interfaces in the reference model are "out
of scope" for the CDNI WG (in the sense that there is no need to
define new protocols for those interfaces), we note that we still need to refer to
them in this document to explain the overall operation of CDNI.We also note that, while we generally show only one Upstream CDN
serving a given CSP, it is entirely possible that multiple uCDNs can
serve a single CSP. In fact, this situation effectively exists today
in the sense that a single CSP can currently delegate its content
delivery to more than one CDN.The following briefly describes the five CDNI interfaces,
paraphrasing the definitions given in . We
discuss these interfaces in more detail in . CDNI Control interface (CI): Operations to bootstrap and
parameterize the other CDNI interfaces, as well as operations to
pre-position, revalidate, and purge both metadata and content. The
latter subset of operations is sometimes collectively called the
"Trigger interface".CDNI Request Routing interface: Operations to determine what
CDN (and optionally what Surrogate within a CDN) is to serve
end-user requests. This interface is actually a logical bundling
of two separate, but related, interfaces: CDNI Footprint & Capabilities Advertisement interface
(FCI): Asynchronous operations to exchange routing information
(e.g., the network footprint and capabilities served by a
given CDN) that enables CDN selection for subsequent user
requests; andCDNI Request Routing Redirection interface (RI):
Synchronous operations to select a delivery CDN (Surrogate)
for a given user request.CDNI Metadata interface (MI): Operations to communicate
metadata that governs how the content is delivered by
interconnected CDNs. Examples of CDNI Metadata include
geo-blocking directives, availability windows, access control
mechanisms, and purge directives. It may include a combination of:
Asynchronous operations to exchange metadata that govern
subsequent user requests for content; andSynchronous operations that govern behavior for a given
user request for content.CDNI Logging interface (LI): Operations that allow
interconnected CDNs to exchange relevant activity logs. It may
include a combination of:Real-time exchanges, suitable for runtime traffic
monitoring; andOffline exchanges, suitable for analytics and billing.The division between the sets of Trigger-based operations in the
CDNI Control interface and the CDNI Metadata interface is somewhat
arbitrary. For both cases, the information passed from the Upstream
CDN to the Downstream CDN can broadly be viewed as metadata that
describes how content is to be managed by the Downstream CDN. For
example, the information conveyed by the CI to pre-position, revalidate, or
purge metadata is similar to the information conveyed by posting
updated metadata via the MI. Even the CI operation to purge content
could be viewed as a metadata update for that content: purge simply
says that the availability window for the named content ends now. The
two interfaces share much in common, so minimally, there will need to
be a consistent data model that spans both.The distinction we draw has to do with what the uCDN knows about
the successful application of the metadata by the dCDN. In the case of
the CI, the Downstream CDN returning a successful status message
guarantees that the operation has been successfully completed; for example,
the content has been purged or pre-positioned. This implies that the
Downstream CDN accepts responsibility for having successfully
completed the requested operation. In contrast, metadata passed
between CDNs via the MI carries no such completion guarantee.
Returning success implies successful receipt of the metadata, but
nothing can be inferred about precisely when the metadata will take
effect in the Downstream CDN, only that it will take effect
eventually. This is because of the challenge in globally synchronizing
updates to metadata with end-user requests that are currently in
progress (or indistinguishable from currently being in progress).
Clearly, a CDN will not be viewed as a trusted peer if
"eventually" often becomes an indefinite period of time,
but the acceptance of responsibility cannot be as crisply defined for
the MI.Finally, there is a practical issue that impacts all of the CDNI
interfaces, and that is whether or not to optimize CDNI for HTTP
Adaptive Streaming (HAS). We highlight specific issues related to
delivering HAS content throughout this document, but for a more
thorough treatment of the topic, see .The remainder of this document is organized as follows: describes some essential building
blocks for CDNI, notably the various options for redirecting user
requests to a given CDN. provides a number of illustrative
examples of various CDNI operations. describes the functionality of the
main CDNI interfaces. shows how various deployment models of
CDNI may be achieved using the defined interfaces. describes the trust model of CDNI and
the issues of transitive trust in particular that CDNI raises.At its core, CDNI requires the redirection of
requests from one CDN to another. For any given request that is
received by an Upstream CDN, it will either respond to the request
directly, or somehow redirect the request to a Downstream CDN. Two
main mechanisms are available for redirecting a request to a
Downstream CDN. The first leverages the DNS name resolution process
and the second uses application-layer redirection mechanisms such as
the HTTP 302 or Real-Time Streaming Protocol (RTSP) 302 redirection responses. While there exists a
large variety of application-layer protocols that include some form of
redirection mechanism, this document will use HTTP (and HTTPS) in its
examples. Similar mechanisms can be applied to other application-layer
protocols. What follows is a short discussion of both DNS- and
HTTP-based redirection, before presenting some examples of their use
in .DNS redirection is based on returning different IP addresses for
the same DNS name, for example, to balance server load or to account
for the client's location in the network. A DNS server,
sometimes called the Local DNS (LDNS), resolves DNS names on behalf
of an end user. The LDNS server in turn queries other DNS servers
until it reaches the authoritative DNS server for the CDN-Domain.
The network operator typically provides the LDNS server, although
the user is free to choose other DNS servers (e.g., OpenDNS, Google
Public DNS). This latter possibility is important because the
authoritative DNS server sees only the IP address of the DNS server
that queries it, not the IP address of the original end user.The advantage of DNS redirection is that it is completely
transparent to the end user; the user sends a DNS name to the LDNS
server and gets back an IP address. On the other hand, DNS
redirection is problematic because the DNS request comes from the
LDNS server, not the end user. This may affect the accuracy of
server selection that is based on the user's location. The
transparency of DNS redirection is also a problem in that there is
no opportunity to take the attributes of the user agent or the URI
path component into account. We consider two main forms of DNS
redirection: simple and CNAME-based.In simple DNS redirection, the authoritative DNS server for the
name simply returns an IP address from a set of possible IP
addresses. The answer is chosen from the set based on
characteristics of the set (e.g., the relative loads on the servers)
or characteristics of the client (e.g., the location of the client
relative to the servers). Simple redirection is straightforward. The
only caveats are (1) there is a limit to the number of alternate IP
addresses a single DNS server can manage; and (2) DNS responses are
cached by Downstream servers so the Time to Live (TTL) on the response must be set
to an appropriate value so as to preserve the freshness of the
redirection.In CNAME-based DNS redirection, the authoritative server returns
a CNAME response to the DNS request, telling the LDNS server to
restart the name lookup using a new name. A CNAME is essentially a
symbolic link in the DNS namespace, and like a symbolic link,
redirection is transparent to the client; the LDNS server gets the
CNAME response and re-executes the lookup. Only when the name has
been resolved to an IP address does it return the result to the
user. Note that DNAME would be preferable to CNAME if it becomes
widely supported.One of the advantages of DNS redirection compared to HTTP
redirection is that it can be cached, reducing load on the
redirecting CDN's DNS server. However, this advantage can also be a
drawback, especially when a given DNS resolver doesn't strictly
adhere to the TTL, which is a known problem in some real-world
environments. In such cases, an end user might end up at a dCDN
without first having passed through the uCDN, which might be an
undesirable scenario from a uCDN point of view.HTTP redirection makes use of the redirection response of the
HTTP protocol (e.g.,"302" or "307"). This
response contains a new URL that the application should fetch
instead of the original URL. By changing the URL appropriately, the
server can cause the user to redirect to a different server. The
advantages of HTTP redirection are that (1) the server can change
the URL fetched by the client to include, for example, both the DNS
name of the particular server to use, as well as the original HTTP
server that was being accessed; (2) the client sends the HTTP
request to the server, so that its IP address is known and can be
used in selecting the server; and (3) other attributes (e.g.,
content type, user agent type) are visible to the redirection
mechanism.Just as is the case for DNS redirection, there are some potential
disadvantages of using HTTP redirection. For example, it may affect
application behavior; web browsers will not send cookies if the
URL changes to a different domain. In addition, although this might
also be an advantage, results of HTTP redirection are not cached so
that all redirections must go through to the uCDN.To provide a big-picture overview of the various components of CDNI, we walk through a "day in the life" of a content item
that is made available via a pair of interconnected CDNs. This will
serve to illustrate many of the functions that need to be supported in a
complete CDNI solution. We give examples using both DNS-based and
HTTP-based redirection. We begin with very simple examples and then show
how additional capabilities, such as recursive request redirection and
content removal, might be added.Before walking through the specific examples, we present a high-level
view of the operations that may take place. This high-level overview is
illustrated in . Note that most
operations will involve only a subset of all the messages shown below,
and that the order and number of operations may vary considerably, as
the more detailed examples illustrate.The following shows Operator A as the Upstream CDN (uCDN) and
Operator B as the Downstream CDN (dCDN), where the former has a
relationship with a content provider and the latter is the CDN
selected by Operator A to deliver content to the end user. The
interconnection relationship may be symmetric between these two CDN
operators, but each direction can be considered as operating
independently of the other; for simplicity, we show the interaction in
one direction only.The operations shown in the figure are as follows: The dCDN uses the FCI to advertise information relevant to its
delivery footprint and capabilities prior to any content requests
being redirected.Prior to any content request, the uCDN uses the MI to
pre-position CDNI Metadata to the dCDN, thereby making that metadata
available in readiness for later content requests.A content request from a user agent arrives at the uCDN.The uCDN may use the RI to synchronously request information from
the dCDN regarding its delivery capabilities to decide if the dCDN is a
suitable target for redirection of this request.The uCDN redirects the request to the dCDN by sending some response (DNS,
HTTP) to the user agent.The user agent requests the content from the dCDN.The dCDN may use the MI to synchronously request metadata related to
this content from uCDN, e.g., to decide whether to serve it.If the content is not already in a suitable cache in the dCDN, the dCDN
may acquire it from the uCDN.The content is delivered to the dCDN from the uCDN.The content is delivered to the user agent by the dCDN.Some time later, perhaps at the request of the CSP (not shown)
the uCDN may use the CI to instruct the dCDN to purge the content, thereby
ensuring it is not delivered again.After one or more content delivery actions by the dCDN, a log of
delivery actions may be provided to the uCDN using the LI.The following sections show some more specific examples of how these
operations may be combined to perform various delivery, control, and
logging operations across a pair of CDNs.Initially, we assume that there is at least one CSP that has
contracted with an Upstream CDN (uCDN) to deliver content on its
behalf. We are not particularly concerned with the interface between
the CSP and uCDN, other than to note that it is expected to be the
same as in the "traditional" (non-interconnected) CDN case. Existing
mechanisms such as DNS CNAMEs or HTTP redirects () can be used to direct a user request for a piece of
content from the CSP towards the CSP's chosen Upstream CDN.We assume Operator A provides an Upstream CDN that serves content
on behalf of a CSP with CDN-Domain cdn.csp.example. We assume that
Operator B provides a Downstream CDN. An end user at some point makes
a request for URLhttp://cdn.csp.example/...rest of URL...It may well be the case that cdn.csp.example is just a CNAME for
some other CDN-Domain (such as csp.op-a.example). Nevertheless, the
HTTP request in the examples that follow is assumed to be for the
example URL above.Our goal is to enable content identified by the above URL to be
served by the CDN of Operator B. In the following sections, we will
walk through some scenarios in which content is served as well as
other CDNI operations such as the removal of content from a Downstream
CDN.In this section, we walk through a simple, illustrative example
using HTTP redirection from a uCDN to a dCDN. The example also assumes the
use of HTTP redirection inside the uCDN and dCDN; however, this is
independent of the choice of redirection approach across CDNs, so an
alternative example could be constructed still showing HTTP
redirection from the uCDN to dCDN but using DNS for the handling of the request
inside each CDN.For this example, we assume that Operators A and B have established
an agreement to interconnect their CDNs, with A being Upstream and B
being Downstream.The operators agree that a CDN-Domain peer-a.op-b.example will be
used as the target of redirections from the uCDN to dCDN. We assume the
name of this domain is communicated by some means to each CDN. (This
could be established out of band or via a CDNI interface.) We refer to
this domain as a "distinguished" CDN-Domain to convey the fact that
its use is limited to the interconnection mechanism; such a domain is
never used directly by a CSP.We assume the operators also agree on some distinguished CDN-Domain
that will be used for inter-CDN acquisition of the CSP's content from the uCDN
by the dCDN. In this example, we'll use op&nbhy;b&nbhy;acq.op&nbhy;a.example.We assume the operators also exchange information regarding which
requests the dCDN is prepared to serve. For example, the dCDN may be prepared
to serve requests from clients in a given geographical region or a set
of IP address prefixes. This information may again be provided out of
band or via a defined CDNI interface.We assume DNS is configured in the following way: The content provider is configured to make Operator A the
authoritative DNS server for cdn.csp.example (or to return a CNAME
for cdn.csp.example for which Operator A is the authoritative DNS
server).Operator A is configured so that a DNS request for
op&nbhy;b&nbhy;acq.op&nbhy;a.example returns a Request Router in Operator A.Operator B is configured so that a DNS request for
peer&nbhy;a.op&nbhy;b.example/cdn.csp.example returns a Request Router in
Operator B. illustrates how a client request
forhttp://cdn.csp.example/...rest of URL...is handled.The steps illustrated in the figure are as follows:A DNS resolver for Operator A processes the DNS request for its
customer based on CDN-Domain cdn.csp.example. It returns the IP
address of a Request Router in Operator A.A Request Router for Operator A processes the HTTP request and
recognizes that the end user is best served by another CDN,
specifically one provided by Operator B, and so it returns a 302
redirect message for a new URL constructed by
"stacking" Operator B's distinguished CDN-Domain
(peer-a.op-b.example) on the front of the original URL. (Note that
more complex URL manipulations are possible, such as replacing the
initial CDN-Domain by some opaque handle.)The end user does a DNS lookup using Operator B's
distinguished CDN-Domain (peer-a.op-b.example). B's DNS
resolver returns the IP address of a Request Router for Operator
B. Note that if request routing within the dCDN was performed using
DNS instead of HTTP redirection, B's DNS resolver would also
behave as the Request Router and directly return the IP address of
a delivery node.The Request Router for Operator B processes the HTTP request
and selects a suitable delivery node to serve the end-user
request, and it returns a 302 redirect message for a new URL
constructed by replacing the hostname with a subdomain of the
Operator B's distinguished CDN-Domain that points to the
selected delivery node.The end user does a DNS lookup using Operator B's
delivery node subdomain (node1.peer-a.op-b.example). B's DNS
resolver returns the IP address of the delivery node.The end user requests the content from B's delivery node.
In the case of a cache hit, steps 6, 7, 8, 9, and 10 below do not
happen, and the content data is directly returned by the delivery
node to the end user. In the case of a cache miss, the content
needs to be acquired by the dCDN from the uCDN (not the CSP). The
distinguished CDN-Domain peer-a.op-b.example indicates to the dCDN
that this content is to be acquired from the uCDN; stripping the
CDN-Domain reveals the original CDN-Domain cdn.csp.example, and
the dCDN may verify that this CDN-Domain belongs to a known peer (so
as to avoid being tricked into serving as an open proxy). It then
does a DNS request for an inter-CDN acquisition CDN-Domain as
agreed above (in this case, op-b-acq.op-a.example).Operator A's DNS resolver processes the DNS request and returns
the IP address of a Request Router in Operator A.The Request Router for Operator A processes the HTTP request
from Operator B's delivery node. Operator A's Request Router
recognizes that the request is from a peer CDN rather than an
end user because of the dedicated inter-CDN acquisition domain
(op&nbhy;b&nbhy;acq.op&nbhy;a.example). (Note that without this specially defined
inter-CDN acquisition domain, Operator A would be at risk of
redirecting the request back to Operator B, resulting in an
infinite loop). The Request Router for Operator A selects a
suitable delivery node in uCDN to serve the inter-CDN acquisition
request and returns a 302 redirect message for a new URL
constructed by replacing the hostname with a subdomain of the
Operator A's distinguished inter-CDN acquisition domain that
points to the selected delivery node.Operator A's DNS resolver processes the DNS request and returns
the IP address of the delivery node in Operator A.Operator B requests (acquires) the content from Operator A.
Although not shown, Operator A processes the rest of the URL: it
extracts information identifying the origin server, validates that
this server has been registered, and determines the content
provider that owns the origin server. It may also perform its own
content acquisition steps if needed before returning the content
to dCDN.The main advantage of this design is that it is simple: each CDN
need only know the distinguished CDN-Domain for each peer, with the
Upstream CDN "pushing" the Downstream CDN-Domain onto the
URL as part of its redirect (step 2), and the Downstream CDN
"popping" its CDN-Domain off the URL to expose a
CDN-Domain that the Upstream CDN can correctly process. Neither CDN
need be aware of the internal structure of the other's URLs.
Moreover, the inter-CDN redirection is entirely supported by a single
HTTP redirect; neither CDN need be aware of the other's internal
redirection mechanism (i.e., whether it is DNS or HTTP based).One disadvantage is that the end user's browser is redirected to a
new URL that is not in the same domain of the original URL. This has
implications on a number of security or validation mechanisms
sometimes used on endpoints. For example, it is important that any
redirected URL be in the same domain (e.g., csp.example) if the
browser is expected to send any cookies associated with that domain.
As another example, some video players enforce validation of a cross-domain policy that needs to accommodate the domains involved in the
CDN redirection. These problems are generally solvable, but the
solutions complicate the example, so we do not discuss them further in
this document.We note that this example begins to illustrate some of the
interfaces that may be required for CDNI, but it does not require all of
them. For example, obtaining information from a dCDN regarding the set
of client IP addresses or geographic regions it might be able to serve
is an aspect of request routing (specifically of the CDNI Footprint
& Capabilities Advertisement interface). Important configuration
information such as the distinguished names used for redirection and
inter-CDN acquisition could also be conveyed via a CDNI interface
(e.g., perhaps the CDNI Control interface). The example also shows how
existing HTTP-based methods suffice for the acquisition interface.
Arguably, the absolute minimum metadata required for CDNI is the
information required to acquire the content, and this information was
provided "in-band" in this example by means of the URI handed to the
client in the HTTP 302 response. The example also assumes that the CSP
does not require any distribution policy (e.g., time window or
geo-blocking) or delivery processing to be applied by the
interconnected CDNs. Hence, there is no explicit CDNI Metadata
interface invoked in this example. There is also no explicit CDNI
Logging interface discussed in this example.We also note that the step of deciding when a request should be
redirected to the dCDN rather than served by the uCDN has been somewhat
glossed over. It may be as simple as checking the client IP address
against a list of prefixes, or it may be considerably more complex,
involving a wide range of factors, such as the geographic location of
the client (perhaps determined from a third-party service), CDN load,
or specific business rules.This example uses the "iterative" CDNI request redirection
approach. That is, a uCDN performs part of the request redirection
function by redirecting the client to a Request Router in the dCDN,
which then performs the rest of the redirection function by
redirecting to a suitable Surrogate. If request routing is performed
in the dCDN using HTTP redirection, this translates in the end user
experiencing two successive HTTP redirections. By contrast, the
alternative approach of "recursive" CDNI request redirection
effectively coalesces these two successive HTTP redirections into a
single one, sending the end user directly to the right delivery node
in the dCDN. This "recursive" CDNI request routing approach is
discussed in the next section.While the example above uses HTTP, the iterative HTTP redirection
mechanism would work over HTTPS in a similar fashion. In order to make
sure an end user's HTTPS request is not downgraded to HTTP along the
redirection path, it is necessary for every Request Router along the
path from the initial uCDN Request Router to the final Surrogate in
the dCDN to respond to an incoming HTTPS request with an HTTP redirect
containing an HTTPS URL. It should be noted that using HTTPS will have
the effect of increasing the total redirection process time and
increasing the load on the Request Routers, especially when the
redirection path includes many redirects and thus many Secure Socket Layer/Transport Layer Security (SSL/TLS)
sessions. In such cases, a recursive HTTP redirection mechanism, as
described in an example in the next section, might help to reduce some
of these issues.The following example builds on the previous one to illustrate the
use of the request routing interface (specifically, the CDNI Request
Routing Redirection interface) to enable "recursive" CDNI request
routing. We build on the HTTP-based redirection approach because it
illustrates the principles and benefits clearly, but it is equally
possible to perform recursive redirection when DNS-based redirection
is employed.In contrast to the prior example, the operators need not agree in
advance on a CDN-Domain to serve as the target of redirections from
the uCDN to dCDN. We assume that the operators agree on some distinguished
CDN-Domain that will be used for inter-CDN acquisition of the CSP's
content by dCDN. In this example, we'll use op&nbhy;b&nbhy;acq.op&nbhy;a.example.We assume the operators also exchange information regarding which
requests the dCDN is prepared to serve. For example, the dCDN may be prepared
to serve requests from clients in a given geographical region or a set
of IP address prefixes. This information may again be provided out of
band or via a defined protocol.We assume DNS is configured in the following way: The content provider is configured to make Operator A the
authoritative DNS server for cdn.csp.example (or to return a CNAME
for cdn.csp.example for which Operator A is the authoritative DNS
server).Operator A is configured so that a DNS request for
op&nbhy;b&nbhy;acq.op&nbhy;a.example returns a Request Router in Operator A.Operator B is configured so that a request for
node1.op&nbhy;b.example/cdn.csp.example returns the IP address of a
delivery node. Note that there might be a number of such delivery
nodes. illustrates how a client request
forhttp://cdn.csp.example/...rest of URL...is handled.The steps illustrated in the figure are as follows:A DNS resolver for Operator A processes the DNS request for its
customer based on CDN-Domain cdn.csp.example. It returns the IP
address of a Request Router in Operator A.A Request Router for Operator A processes the HTTP request and
recognizes that the end user is best served by another
CDN -- specifically one provided by Operator B -- so it
queries the CDNI Request Routing Redirection interface of Operator
B, providing a set of information about the request including the
URL requested. Operator B replies with the DNS name of a delivery
node.Operator A returns a 302 redirect message for a new URL
obtained from the RI.The end user does a DNS lookup using the hostname of the URL
just provided (node1.op-b.example). B's DNS resolver returns
the IP address of the corresponding delivery node. Note that,
since the name of the delivery node was already obtained from B
using the RI, there should not be any further redirection here (in
contrast to the iterative method described above.)The end user requests the content from B's delivery node,
potentially resulting in a cache miss. In the case of a cache
miss, the content needs to be acquired from the uCDN (not the CSP.)
The distinguished CDN-Domain op-b.example indicates to the dCDN that
this content is to be acquired from another CDN; stripping the
CDN-Domain reveals the original CDN-Domain cdn.csp.example. The dCDN
may verify that this CDN-Domain belongs to a known peer (so as to
avoid being tricked into serving as an open proxy). It then does a
DNS request for the inter-CDN Acquisition
"distinguished" CDN-Domain as agreed above (in this
case, op-b-acq.op-a.example).Operator A's DNS resolver processes the DNS request and returns
the IP address of a Request Router in Operator A.The Request Router for Operator A processes the HTTP request
from Operator B's delivery node. Operator A's Request Router
recognizes that the request is from a peer CDN rather than an
end user because of the dedicated inter-CDN acquisition domain
(op&nbhy;b&nbhy;acq.op&nbhy;a.example). (Note that without this specially defined
inter-CDN acquisition domain, Operator A would be at risk of
redirecting the request back to Operator B, resulting in an
infinite loop). The Request Router for Operator A selects a
suitable delivery node in the uCDN to serve the inter-CDN acquisition
request and returns a 302 redirect message for a new URL
constructed by replacing the hostname with a subdomain of the
Operator A's distinguished inter-CDN acquisition domain that
points to the selected delivery node.Operator A recognizes that the DNS request is from a peer CDN
rather than an end user (due to the internal CDN-Domain) and so
returns the address of a delivery node. (Note that without this
specially defined internal domain, Operator A would be at risk of
redirecting the request back to Operator B, resulting in an
infinite loop.)Operator B requests (acquires) the content from Operator A.
Operator A serves content for the requested CDN-Domain to the dCDN.
Although not shown, it is at this point that Operator A processes
the rest of the URL: it extracts information identifying the
origin server, validates that this server has been registered, and
determines the content provider that owns the origin server. It
may also perform its own content acquisition steps if needed
before returning the content to the dCDN.Recursive redirection has the advantage (over iterative redirection) of being
more transparent from the end user's perspective but the disadvantage
of each CDN exposing more of its internal structure (in particular,
the addresses of edge caches) to peer CDNs. By contrast, iterative
redirection does not require the dCDN to expose the addresses of its edge
caches to the uCDN.This example happens to use HTTP-based redirection in both CDN A
and CDN B, but a similar example could be constructed using DNS-based
redirection in either CDN. Hence, the key point to take away here is
simply that the end user only sees a single redirection of some type,
as opposed to the pair of redirections in the prior (iterative)
example.The use of the RI requires that the request routing mechanism be
appropriately configured and bootstrapped, which is not shown here.
More discussion on the bootstrapping of interfaces is provided in
In this section we walk through a simple example using DNS-based
redirection for request redirection from the uCDN to the dCDN (as well as for
request routing inside the dCDN and the uCDN). As noted in , DNS-based redirection has certain advantages
over HTTP-based redirection (notably, it is transparent to the
end user) as well as some drawbacks (notably, the client IP address is
not visible to the Request Router).As before, Operator A has to learn the set of requests that the dCDN is
willing or able to serve (e.g., which client IP address prefixes or
geographic regions are part of the dCDN footprint). We assume Operator
B has and makes known to Operator A some unique identifier that can be
used for the construction of a distinguished CDN-Domain, as shown in
more detail below. (This identifier strictly needs only to be unique
within the scope of Operator A, but a globally unique identifier, such
as an Autonomous System (AS) number assigned to B, is one easy way to achieve that.) Also,
Operator A obtains the NS records for Operator B's externally visible
redirection servers. Also, as before, a distinguished CDN-Domain, such
as op&nbhy;b&nbhy;acq.op&nbhy;a.example, must be assigned for inter-CDN
acquisition.We assume DNS is configured in the following way: The CSP is configured to make Operator A the authoritative DNS
server for cdn.csp.example (or to return a CNAME for
cdn.csp.example for which Operator A is the authoritative DNS
server).When uCDN sees a request best served by the dCDN, it returns CNAME
and NS records for "b.cdn.csp.example", where "b" is the unique
identifier assigned to Operator B. (It may, for example, be an AS
number assigned to Operator B.)The dCDN is configured so that a request for "b.cdn.csp.example"
returns a delivery node in the dCDN.The uCDN is configured so that a request for
"op&nbhy;b&nbhy;acq.op&nbhy;a.example" returns a delivery node in the uCDN. depicts the exchange of DNS and HTTP
requests. The main differences from are
the lack of HTTP redirection and transparency to the end user.The steps illustrated in the figure are as follows:The Request Router for Operator A processes the DNS request for
CDN-Domain cdn.csp.example and recognizes that the end user is
best served by another CDN. (This may depend on the IP address of
the user's LDNS resolver, or other information discussed
below.) The Request Router returns a DNS CNAME response by
"stacking" the distinguished identifier for Operator B
onto the original CDN-Domain (e.g., b.cdn.csp.example).The end user sends a DNS query for the modified CDN-Domain
(i.e., b.cdn.csp.example) to Operator A's DNS server. The Request
Router for Operator A processes the DNS request and returns a
delegation to b.cdn.csp.example by sending an NS record plus glue
records pointing to Operator B's DNS server. (This extra
step is necessary since typical DNS implementation won't follow an
NS record when it is sent together with a CNAME record, thereby
necessitating a two-step approach.)The end user sends a DNS query for the modified CDN-Domain
(i.e., b.cdn.csp.example) to Operator B's DNS server, using the NS
and AAAA/A records received in step 2. This causes B's
Request Router to respond with a suitable delivery node.The end user requests the content from B's delivery node.
The requested URL contains the name cdn.csp.example. (Note that
the returned CNAME does not affect the URL.) At this point, the
delivery node has the correct IP address of the end user and can
do an HTTP 302 redirect if the redirections in steps 2 and 3 were
incorrect. Otherwise, B verifies that this CDN-Domain belongs to a
known peer (so as to avoid being tricked into serving as an open
proxy). It then does a DNS request for an "internal"
CDN-Domain as agreed above (op-b-acq.op-a.example).Operator A recognizes that the DNS request is from a peer CDN
rather than an end user (due to the internal CDN-Domain) and so
returns the address of a delivery node in uCDN.Operator A serves content to dCDN. Although not shown, it is at
this point that Operator A processes the rest of the URL: it
extracts information identifying the origin server, validates that
this server has been registered, and determines the content
provider that owns the origin server.The advantages of this approach are that it is more transparent to
the end user and requires fewer round trips than HTTP-based
redirection (in its worst case, i.e., when none of the needed DNS
information is cached). A potential problem is that the Upstream CDN
depends on being able to learn the correct Downstream CDN that serves
the end user from the client address in the DNS request. In standard
DNS operation, the uCDN will only obtain the address of the client's LDNS resolver, which is not guaranteed to be in the same network
(or geographic region) as the client. If not (e.g., the end user
uses a global DNS service), then the Upstream CDN cannot determine
the appropriate Downstream CDN to serve the end user. In this case,
and assuming the uCDN is capable of detecting that situation, one
option is for the Upstream CDN to treat the end user as it would any
user not connected to a peer CDN. Another option is for the Upstream
CDN to "fall back" to a pure HTTP-based redirection
strategy in this case (i.e., use the first method). Note that this
problem affects existing CDNs that rely on DNS to determine where to
redirect client requests, but the consequences are arguably less
serious for CDNI since the LDNS is likely in the same network as the
dCDN serves.As with the prior example, this example partially illustrates the
various interfaces involved in CDNI. Operator A could learn
dynamically from Operator B the set of prefixes or regions that B is
willing and able to serve via the CDNI Footprint & Capabilities
Advertisement interface. The distinguished name used for acquisition
and the identifier for Operator B that is prepended to the CDN-Domain
on redirection are examples of information elements that might also be
conveyed by CDNI interfaces (or, alternatively, statically
configured). As before, minimal metadata sufficient to obtain the
content is carried "in-band" as part of the redirection process, and
standard HTTP is used for inter-CDN acquisition. There is no explicit
CDNI Logging interface discussed in this example.Although it is possible to use DNSSEC in combination with the
Iterative DNS-based Redirection mechanism explained above, it is
important to note that the uCDN might have to sign records on the
fly, since the CNAME returned, and thus the signature provided, can
potentially be different for each incoming query. Although there is
nothing preventing a uCDN from performing such on-the-fly signing,
this might be computationally expensive. In the case where the
number of dCDNs, and thus the number of different CNAMEs to return,
is relatively stable, an alternative solution would be for the uCDN
to pre-generate signatures for all possible CNAMEs. For each
incoming query, the uCDN would then determine the appropriate CNAME
and return it together with the associated pre-generated signature.
Note: In the latter case, maintaining the serial number and signature of Start of Authority (SOA)
might be an issue since, technically, it should change every time a
different CNAME is used. However, since, in practice, direct SOA
queries are relatively rare, a uCDN could defer incrementing the
serial number and resigning the SOA until it is queried and then do it
on-the-fly.Note also that the NS record and the glue records used in
step 2 in the previous section should generally be identical to
those of their authoritative zone managed by Operator B. Even if
they differ, this will not make the DNS resolution process fail, but
the client DNS server will prefer the authoritative data in its
cache and use it for subsequent queries. Such inconsistency is a
general operational issue of DNS, but it may be more important for
this architecture because the uCDN (Operator A) would rely on the
consistency to make the resulting redirection work as intended. In
general, it is the administrator's responsibility to make them
consistent.There could be situations where being able to dynamically discover
the set of requests that a given dCDN is willing and able to serve is
beneficial. For example, a CDN might at one time be able to serve a
certain set of client IP prefixes, but that set might change over time
due to changes in the topology and routing policies of the IP network.
The following example illustrates this capability. We have chosen the
example of DNS-based redirection, but HTTP-based redirection could
equally well use this approach.This example differs from the one in
only in the addition of an RI request (step 2) and corresponding
response (step 3). The RI REQ could be a message such as "Can you
serve clients from this IP Prefix?" or it could be "Provide the list
of client IP prefixes you can currently serve". In either case the
response might be cached by Operator A to avoid repeatedly asking the
same question. Alternatively, or in addition, Operator B may
spontaneously advertise to Operator A information (or changes) on the
set of requests it is willing and able to serve on behalf of Operator
A; in that case, Operator B may spontaneously issue RR/RI REPLY
messages that are not in direct response to a corresponding RR/RI REQ
message. (Note that the issues of determining the client's subnet from
DNS requests, as described above, are exactly the same here as in
.)Once Operator A obtains the RI response, it is now able to
determine that Operator B's CDN is an appropriate dCDN for this
request and therefore a valid candidate dCDN to consider in its
redirection decision. If that dCDN is selected, the redirection and
serving of the request proceeds as before (i.e., in the absence of
dynamic footprint discovery).The following example illustrates how the CDNI Control interface
may be used to achieve pre-positioning of an item of content in the
dCDN. In this example, user requests for a particular content, and
corresponding redirection of such requests from Operator A to Operator
B CDN, may (or may not) have taken place earlier. Then, at some point
in time, the uCDN (for example, in response to a corresponding Trigger
from the Content Provider) uses the CI to request that content
identified by a particular URL be removed from dCDN. The following
diagram illustrates the operation. It should be noted that a uCDN will
typically not know whether a dCDN has cached a given content item;
however, it may send the content removal request to make sure no
cached versions remain to satisfy any contractual obligations it may
have.The CI is used to convey the request from the uCDN to the dCDN that some
previously acquired content should be deleted. The URL in the request
specifies which content to remove. This example corresponds to a
DNS-based redirection scenario such as . If
HTTP-based redirection had been used, the URL for removal would be of
the form peer-a.op-b.example/cdn.csp.example/...The dCDN is expected to confirm to the uCDN, as illustrated by the
CI OK message, the completion of the removal of the targeted content
from all the caches in the dCDN.The following example illustrates how the CI may be used to
pre-position an item of content in the dCDN. In this example, Operator
A uses the CDNI Metadata interface to request that content identified
by a particular URL be pre-positioned into Operator B CDN.The steps illustrated in the figure are as follows:Operator A uses the CI to request that Operator B pre-position
a particular content item identified by its URL. Operator B
responds by confirming that it is willing to perform this
operation.Steps 2 and 3 are exactly the same as steps 5 and 6 of , only this time those steps happen as the
result of the Pre-positioning request instead of as the result of a
cache miss.Steps 4, 5, 6, and 7 are exactly the same as steps 1, 2, 3, and 4 of , only this time, Operator B's CDN can serve the
end-user request without triggering dynamic content acquisition, since
the content has been pre-positioned in the dCDN. Note that, depending on
dCDN operations and policies, the content pre-positioned in the dCDN
may be pre-positioned to all, or a subset of, dCDN caches. In the
latter case, intra-CDN dynamic content acquisition may take place
inside the dCDN serving requests from caches on which the content has
not been pre-positioned; however, such intra-CDN dynamic acquisition
would not involve the uCDN.In this section, we walk through a simple example illustrating a
scenario of asynchronously exchanging CDNI Metadata, where the
Downstream CDN obtains CDNI Metadata for content ahead of a
corresponding content request. The example that follows assumes that
HTTP-based inter-CDN redirection and recursive CDNI request routing
are used, as in . However, Asynchronous
exchange of CDNI Metadata is similarly applicable to DNS-based
inter-CDN redirection and iterative request routing (in which cases
the CDNI Metadata may be used at slightly different processing stages
of the message flows).The steps illustrated in the figure are as follows:Operator A uses the CI to trigger the signaling of the availability of
CDNI Metadata to Operator B.Operator B acknowledges the receipt of this Trigger.Operator B requests the latest metadata from Operator A using
the MI.Operator A replies with the requested metadata. This document
does not constrain how the CDNI Metadata information is actually
represented. For the purposes of this example, we assume that
Operator A provides CDNI Metadata to Operator B indicating that:
this CDNI Metadata is applicable to any content referenced
by some CDN-Domain.this CDNI Metadata consists of a distribution policy
requiring enforcement by the delivery node of a specific
per-request authorization mechanism (e.g., URI signature or
token validation).A Content Request occurs as usual.A CDNI Request Routing Redirection request (RI REQ) is issued
by Operator A's CDN, as discussed in .
Operator B's Request Router can access the CDNI Metadata that are
relevant to the requested content and that have been
pre-positioned as per Steps 1-4, which may or may not affect the
response.Operator B's Request Router issues a CDNI Request Routing
Redirection response (RI RESP) as in .Operator B performs content redirection as discussed in .On receipt of the Content Request by the end user, the delivery
node detects that previously acquired CDNI Metadata is applicable
to the requested content. In accordance with the specific CDNI
metadata of this example, the delivery node will invoke the
appropriate per-request authorization mechanism, before serving
the content. (Details of this authorization are not shown.)Assuming successful per-request authorization, serving of
Content Data (possibly preceded by inter-CDN acquisition) proceeds
as in .In this section we walk through a simple example illustrating a
scenario of Synchronous CDNI Metadata acquisition, in which the
Downstream CDN obtains CDNI Metadata for content at the time of
handling a first request for the corresponding content. As in the
preceding section, this example assumes that HTTP-based inter-CDN
redirection and recursive CDNI request routing are used (as in ), but dynamic CDNI Metadata acquisition is
applicable to other variations of request routing.The steps illustrated in the figure are as follows:A Content Request arrives as normal.An RI request occurs as in the prior example.On receipt of the CDNI Request Routing Request, Operator B's
CDN initiates Synchronous acquisition of CDNI Metadata that are
needed for routing of the end-user request. We assume the URI for
the a metadata server is known ahead of time through some
out-of-band means.On receipt of a CDNI Metadata request, Operator A's CDN
responds, making the corresponding CDNI Metadata information
available to Operator B's CDN. This metadata is considered by
Operator B's CDN before responding to the Request Routing request.
(In a simple case, the metadata could simply be an allow or deny
response for this particular request.)Response to the RI request as normal.Redirection message is sent to the end user.A delivery node of Operator B receives the end user
request.The delivery node Triggers dynamic acquisition of additional
CDNI Metadata that are needed to process the end-user content
request. Note that there may exist cases where this step need not
happen, for example, because the metadata were already acquired
previously.Operator A's CDN responds to the CDNI Metadata Request and
makes the corresponding CDNI Metadata available to Operator B.
This metadata influence how Operator B's CDN processes the
end-user request.Content is served (possibly preceded by inter-CDN acquisition)
as in .A single dCDN may receive end-user requests from multiple uCDNs.
When a dCDN receives an end-user request, it must determine the
identity of the uCDN from which it should acquire the requested
content.Ideally, the acquisition path of an end-user request will follow
the redirection path of the request. The dCDN should acquire the
content from the same uCDN that redirected the request.Determining the acquisition path requires the dCDN to reconstruct
the redirection path based on information in the end-user request. The
method for reconstructing the redirection path differs based on the
redirection approach: HTTP or DNS.With HTTP-redirection, the rewritten URI should include sufficient
information for the dCDN to directly or indirectly determine the uCDN
when the end-user request is received. The HTTP-redirection approach
can be further broken-down based on the how the URL is rewritten
during redirection: HTTP redirection with or without Site Aggregation.
HTTP redirection with Site Aggregation hides the identity of the
original CSP. HTTP redirection without Site Aggregation does not
attempt to hide the identity of the original CSP. With both
approaches, the rewritten URI includes enough information to identify
the immediate neighbor uCDN.With DNS-redirection, the dCDN receives the published URI (instead
of a rewritten URI) and does not have sufficient information for the
dCDN to identify the appropriate uCDN. The dCDN may narrow the set of
viable uCDNs by examining the CDNI Metadata from each to determine
which uCDNs are hosting metadata for the requested content. If there
is a single uCDN hosting metadata for the requested content, the dCDN
can assume that the request redirection is coming from this uCDN and
can acquire content from that uCDN. If there are multiple uCDNs
hosting metadata for the requested content, the dCDN may be ready to
trust any of these uCDNs to acquire the content (provided the uCDN is
in a position to serve it). If the dCDN is not ready to trust any of
these uCDNs, it needs to ensure via out of band arrangements that, for
a given content, only a single uCDN will ever redirect requests to the
dCDN.Content acquisition may be preceded by content metadata
acquisition. If possible, the acquisition path for metadata should
also follow the redirection path. Additionally, we assume metadata is
indexed based on rewritten URIs in the case of HTTP redirection and is
indexed based on published URIs in the case of DNS-redirection. Thus,
the RI and the MI are tightly coupled in that the result of request
routing (a rewritten URI pointing to the dCDN) serves as an input to
metadata lookup. If the content metadata includes information for
acquiring the content, then the MI is also tightly coupled with the
acquisition interface in that the result of the metadata lookup (an
acquisition URL likely hosted by the uCDN) should serve as input to
the content acquisition. illustrates the main interfaces that are in
scope for the CDNI WG, along with several others. The detailed
specifications of these interfaces are left to other documents, but see
and
for some discussion of the interfaces.One interface that is not shown in is the
interface between the user and the CSP. While for the purposes of CDNI
that interface is out of scope, it is worth noting that it does exist
and can provide useful functions, such as end-to-end performance
monitoring and some forms of authentication and authorization.There is also an important interface between the user and the Request
Routing function of both uCDN and dCDN (shown as the "Request" interface
in ). As we saw in some of the preceding examples, that
interface can be used as a way of passing metadata, such as the minimum
information that is required for dCDN to obtain the content from
the uCDN.In this section we will provide an overview of the functions
performed by each of the CDNI interfaces and discuss how they fit into
the overall solution. We also examine some of the design trade-offs, and
explore several cross-interface concerns. We begin with an examination
of one such trade-off that affects all the interfaces -- the use of
in-band or out-of-band communication.Before getting to the individual interfaces, we observe that there
is a high-level design choice for each, involving the use of existing
in-band communication channels versus defining new out-of-band
interfaces.It is possible that the information needed to carry out various
interconnection functions can be communicated between peer CDNs using
existing in-band protocols. The use of HTTP 302 redirect is an example
of how certain aspects of request routing can be implemented in-band
(embedded in URIs). Note that using existing in-band protocols does
not imply that the CDNI interfaces are null; it is still necessary to
establish the rules (conventions) by which such protocols are used to
implement the various interface functions.There are other opportunities for in-band communication beyond HTTP
redirects. For example, many of the HTTP directives used by proxy
servers can also be used by peer CDNs to inform each other of caching
activity. Of these, one that is particularly relevant is the
If&nbhy;Modified&nbhy;Since directive, which is used with the GET method to make
it conditional: if the requested object has not been modified since
the time specified in this field, a copy of the object will not be
returned, and instead, a 304 (not modified) response will be
returned.Although the CDNI interfaces are largely independent, there are a
set of conventions practiced consistently across all interfaces. Most
important among these is how resources are named, for example, how
the CDNI Metadata and Control interfaces identify the set of resources
to which a given directive applies or the CDNI Logging interface
identifies the set of resources for which a summary record
applies.While, theoretically, the CDNI interfaces could explicitly identify
every individual resource, in practice, they name resource aggregates
(sets of URIs) that are to be treated in a similar way. For example,
URI aggregates can be identified by a CDN-Domain (i.e., the FQDN at
the beginning of a URI) or by a URI-Filter (i.e., a regular expression
that matches a subset of URIs contained in some CDN-Domain). In other
words, CDN-Domains and URI-Filters provide a uniform means to
aggregate sets (and subsets) of URIs for the purpose of defining the
scope for some operation in one of the CDNI interfaces.The Request Routing interface comprises two parts: the Asynchronous
interface used by a dCDN to advertise footprint and capabilities
(denoted FCI) to a uCDN, allowing the uCDN to decide whether to
redirect particular user requests to that dCDN; and the Synchronous
interface used by the uCDN to redirect a user request to the dCDN
(denoted RI). (These are somewhat analogous to the operations of
routing and forwarding in IP.)As illustrated in , the RI part of
request routing may be implemented in part by DNS and HTTP. Naming
conventions may be established by which CDN peers communicate whether
a request should be routed or content served.We also note that RI plays a key role in enabling recursive
redirection, as illustrated in . It enables
the user to be redirected to the correct delivery node in dCDN with
only a single redirection step (as seen by the user). This may be
particularly valuable as the chain of interconnected CDNs increases
beyond two CDNs. For further discussion on the RI, see .In support of these redirection requests, it is necessary for CDN
peers to exchange additional information with each other, and this is
the role of the FCI part of request routing. Depending on the
method(s) supported, this might include: The operator's unique id (operator-id) or distinguished
CDN-Domain (operator-domain);NS records for the operator's set of externally visible
Request Routers;The set of requests the dCDN operator is prepared to serve
(e.g., a set of client IP prefixes or geographic regions that may
be served by the dCDN).Additional capabilities of the dCDN, such as its ability to
support different CDNI Metadata requests.Note that the set of requests that a dCDN is willing to serve could
in some cases be relatively static (e.g., a set of IP prefixes),
could be exchanged off-line, or might even be negotiated as part of a
peering agreement. However, it may also be more dynamic, in which case
the exchange supported by FCI would be helpful. A further
discussion of the Footprint & Capability Advertisement interface
can be found in .It is necessary for the Upstream CDN to have visibility into the
delivery of content that it redirected to a Downstream CDN. This
allows the Upstream CDN to properly bill its customers for multiple
deliveries of content cached by the Downstream CDN, as well as to
report accurate traffic statistics to those content providers. This is
one role of the LI.Other operational data that may be relevant to CDNI can also be
exchanged by the LI. For example, a dCDN may report the amount of
content it has acquired from uCDN, and how much cache storage has been
consumed by content cached on behalf of uCDN.Traffic logs are easily exchanged off-line. For example, the
following traffic log is a small deviation from the Apache log file
format, where entries include the following fields: Domain - the full domain name of the
origin serverIP address - the IP address of the
client making the requestEnd time - the ending time of the
transferTime zone - any time zone modifier for
the end timeMethod - the transfer command itself
(e.g., GET, POST, HEAD)URL - the requested URLVersion - the protocol version, such
as HTTP/1.0Response - a numeric response code
indicating transfer resultBytes Sent - the number of bytes in
the body sent to the clientRequest ID - a unique identifier for
this transferUser agent - the user agent, if
suppliedDuration - the duration of the
transfer in millisecondsCached Bytes - the number of body
bytes served from the cacheReferrer - the referrer string from the
client, if suppliedOf these, only the Domain field is indirect in the Downstream
CDN -- it is set to the CDN-Domain used by the Upstream CDN rather
than the actual origin server. This field could then used to filter
traffic log entries so only those entries matching the Upstream CDN
are reported to the corresponding operator. Further discussion of the
LI can be found in .One open question is who does the filtering. One option is that the
Downstream CDN filters its own logs and passes the relevant records
directly to each Upstream peer. This requires that the Downstream CDN
know the set of CDN-Domains that belong to each Upstream peer. If
this information is already exchanged between peers as part of another
interface, then direct peer-to-peer reporting is straightforward. If
it is not available, and operators do not wish to advertise the set of
CDN-Domains they serve to their peers, then the second option is for
each CDN to send both its non-local traffic records and the set of
CDN-Domains it serves to an independent third party (i.e., a CDN
Exchange), which subsequently filters, merges, and distributes traffic
records on behalf of each participating CDN operator.A second open question is how timely traffic information should be.
For example, in addition to offline traffic logs, accurate real-time
traffic monitoring might also be useful, but such information requires
that the Downstream CDN inform the Upstream CDN each time it serves
Upstream content from its cache. The Downstream CDN can do this, for
example, by sending a conditional HTTP GET request (If&nbhy;Modified&nbhy;Since)
to the Upstream CDN each time it receives an HTTP GET request from one
of its end users. This allows the Upstream CDN to record that a
request has been issued for the purpose of real-time traffic
monitoring. The Upstream CDN can also use this information to validate
the traffic logs received later from the Downstream CDN.There is obviously a trade-off between accuracy of such monitoring
and the overhead of the Downstream CDN having to go back to the
Upstream CDN for every request.Another design trade-off in the LI is the degree of aggregation or
summarization of data. One situation that lends itself to
summarization is the delivery of HTTP Adaptive Streaming (HAS), since
the large number of individual chunk requests potentially results in
large volumes of logging information. This case is discussed below,
but other forms of aggregation may also be useful. For example, there
may be situations where bulk metrics such as bytes delivered per hour
may suffice rather than the detailed per-request logs outlined above.
It seems likely that a range of granularities of logging will be
needed along with ways to specify the type and degree of aggregation
required.The CDNI Control interface is initially used to bootstrap the other
interfaces. As a simple example, it could be used to provide the
address of the logging server in the dCDN to the uCDN in order to bootstrap
the CDNI Logging interface. It may also be used, for example, to
establish security associations for the other interfaces.The other role the CI plays is to allow the uCDN to pre-position,
revalidate, or purge metadata and content on a dCDN. These operations,
sometimes collectively called the "Trigger interface", are discussed
further in .The role of the CDNI Metadata interface is to enable CDNI
distribution metadata to be conveyed to the Downstream CDN by the
Upstream CDN. Such metadata includes geo-blocking restrictions,
availability windows, access-control policies, and so on. It may also
include information to facilitate acquisition of content by a dCDN
(e.g., alternate sources for the content, authorization information
needed to acquire the content from the source). For a full discussion
of the CDNI Metadata interface, see Some distribution metadata may be partially emulated using in-band
mechanisms. For example, in case of any geo-blocking restrictions or
availability windows, the Upstream CDN can elect to redirect a request
to the Downstream CDN only if that CDN's advertised delivery footprint
is acceptable for the requested URL. Similarly, the request could be
forwarded only if the current time is within the availability window.
However, such approaches typically come with shortcomings such as
inability to prevent from replay outside the time window or inability
to make use of a Downstream CDN that covers a broader footprint than
the geo-blocking restrictions.Similarly, some forms of access control may also be performed on a
per-request basis using HTTP directives. For example, being able to
respond to a conditional GET request gives the Upstream CDN an
opportunity to influence how the Downstream CDN delivers its content.
Minimally, the Upstream CDN can invalidate (purge) content previously
cached by the Downstream CDN.All of these in-band techniques serve to illustrate that uCDNs have
the option of enforcing some of their access control policies
themselves (at the expense of increased inter-CDN signaling load),
rather than delegating enforcement to dCDNs using the MI. As a
consequence, the MI could provide a means for the uCDN to express its
desire to retain enforcement for itself. For example, this might be
done by including a "check with me" flag in the metadata
associated with certain content. The realization of such in-band
techniques over the various inter-CDN acquisition protocols (e.g.,
HTTP) requires further investigation and may require small extensions
or semantic changes to the acquisition protocol.We consider HTTP Adaptive Streaming (HAS) and the impact it has on
the CDNI interfaces because large objects (e.g., videos) are broken
into a sequence of small, independent chunks. For each of the
following, a more thorough discussion, including an overview of the
trade-offs involved in alternative designs, can be found in RFC
6983.First, with respect to Content Acquisition and File Management,
which are out of scope for the CDNI interfaces but, nonetheless, relevant
to the overall operation, we assume no additional measures are
required to deal with large numbers of chunks. This means that the
dCDN is not explicitly made aware of any relationship between
different chunks, and the dCDN handles each chunk as if it were an
individual and independent content item. The result is that content
acquisition between uCDN and dCDN also happens on a per-chunk basis.
This approach is in line with the recommendations made in RFC 6983,
which also identifies potential improvements in this area that might
be considered in the future.Second, with respect to request routing, we note that HAS manifest
files have the potential to interfere with request routing since
manifest files contain URLs pointing to the location of content
chunks. To make sure that a manifest file does not hinder CDNI request
routing and does not place excessive load on CDNI resources, either the use
of manifest files could be limited to those containing relative
URLs or the uCDN could modify the URLs in the manifest. Our approach
for dealing with these issues is twofold. As a mandatory requirement,
CDNs should be able to handle unmodified manifest files containing
either relative or absolute URLs. To limit the number of redirects,
and thus the load placed on the CDNI interfaces, as an optional
feature uCDNs can use the information obtained through the CDNI
Request Routing Redirection interface to modify the URLs in the
manifest file. Since the modification of the manifest file is an
optional uCDN-internal process, this does not require any
standardization effort beyond being able to communicate chunk
locations in the CDNI Request Routing Redirection interface.Third, with respect to the CDNI Logging interface, there are
several potential issues, including the large number of individual
chunk requests potentially resulting in large volumes of logging
information, and the desire to correlate logging information for chunk
requests that correspond to the same HAS session. For the initial CDNI
specification, our approach is to expect participating CDNs to support
per-chunk logging (e.g., logging each chunk request as if it were an
independent content request) over the CDNI Logging interface.
Optionally, the LI may include a Content Collection IDentifier (CCID)
and/or a Session IDentifier (SID) as part of the logging fields,
thereby facilitating correlation of per-chunk logs into per-session
logs for applications benefiting from such session level information
(e.g., session-based analytics). This approach is in line with the
recommendations made in RFC 6983, which also identifies potential
improvements in this area that might be considered in the future.Fourth, with respect to the CDNI Control interface, and in
particular purging HAS chunks from a given CDN, our approach is to
expect each CDN supports per-chunk content purge (e.g., purging of
chunks as if they were individual content items). Optionally, a CDN
may support content purge on the basis of a "Purge IDentifier
(Purge-ID)" allowing the removal of all chunks related to a given
Content Collection with a single reference. It is possible that this
Purge-ID could be merged with the CCID discussed above for HAS
Logging, or alternatively, they may remain distinct.When using HTTP redirection, content URIs may be rewritten when
redirection takes place within a uCDN, from a uCDN to a dCDN, and
within the dCDN. In the case of cascaded CDNs, content URIs may be
rewritten at every CDN hop (e.g., between the uCDN and the dCDN acting
as the transit CDN, and between the transit CDN and the dCDN serving
the request). The content URI used between any uCDN/dCDN pair becomes a
common handle that can be referred to without ambiguity by both CDNs
in all their inter-CDN communications. This handle allows the uCDN and
dCDN to correlate information exchanged using other CDNI interfaces in
both the Downstream direction (e.g., when using the MI) and the
Upstream direction (e.g., when using the LI).Consider the simple case of a single uCDN/dCDN pair using HTTP
redirection. We introduce the following terminology for content URIs
to simplify the discussion:"u-URI" represents a content URI in a request presented to the
uCDN;"ud-URI" is a content URI acting as the common handle across
uCDN and dCDN for requests redirected by the uCDN to a specific
dCDN;"d-URI" represents a content URI in a request made within the
delegate dCDN.In our simple pair-wise example, the "ud-URI" effectively becomes
the handle that the uCDN/dCDN pair use to correlate all CDNI
information. In particular, for a given pair of CDNs executing the
HTTP redirection, the uCDN needs to map the u-URI to the ud-URI handle
for all MI message exchanges, while the dCDN needs to map the d-URI to
the ud-URI handle for all LI message exchanges.In the case of cascaded CDNs, the transit CDN will rewrite the
content URI when redirecting to the dCDN, thereby establishing a new
handle between the transit CDN and the dCDN, that is different from
the handle between the uCDN and transit CDN. It is the responsibility
of the transit CDN to manage its mapping across handles so the right
handle for all pairs of CDNs is always used in its CDNI
communication.In summary, all CDNI interfaces between a given pair of CDNs need
to always use the "ud-URI" handle for that specific CDN pair as their
content URI reference.In this section, we describe a number of possible deployment models
that may be achieved using the CDNI interfaces described above. We note
that these models are by no means exhaustive and that many other models
may be possible.Although the reference model of shows all CDN
functions on each side of the CDNI interface, deployments can rely on
entities that are involved in any subset of these functions, and
therefore only support the relevant subset of CDNI interfaces. As
already noted in , effective CDNI deployments
can be built without necessarily implementing all the interfaces. Some
examples of such deployments are shown below.Note that, while we refer to Upstream and Downstream CDNs, this
distinction applies to specific content items and transactions. That is,
a given CDN may be Upstream for some transactions and Downstream for
others, depending on many factors such as location of the requesting
client and the particular piece of content requested.Although the reference model illustrated in
shows a unidirectional CDN interconnection with a single uCDN and a
single dCDN, any arbitrary CDNI meshing can be built from this, such
as the example meshing illustrated in .
(Support for arbitrary meshing may or may not be in the initial scope
for the working group, but the model allows for it.)Note that our terminology refers to functional roles and not
economic or business roles. That is, a given organization may be
operating as both a CSP and a fully fledged uCDN when we consider the
functions performed, as illustrated in .As another example, a content provider organization may choose to
run its own Request Routing function as a way to select among multiple
candidate CDN providers; in this case, the content provider may be
modeled as the combination of a CSP and of a special, restricted case
of a CDN. In that case, as illustrated in , the CDNI Request Routing interfaces can be
used between the restricted CDN operated by the content provider
Organization and the CDN operated by the full CDN organization acting
as a dCDN in the request routing control plane. Interfaces outside the
scope of the CDNI work can be used between the CSP functional entities
of the content provider organization and the CDN operated by the full
CDN organization acting as a uCDN) in the CDNI control planes other
than the request routing plane (i.e., Control, Distribution,
Logging).There are two additional concepts related to, but distinct from, CDNI. The first is CDN Federation. Our view is that CDNI is
the more general concept, involving two or more CDNs serving content
to each other's users, while federation implies a multi-lateral
interconnection arrangement, but other CDNI agreements
are also possible (e.g., symmetric bilateral, asymmetric bilateral).
An important conclusion is that CDNI technology should not presume (or
bake in) a particular interconnection agreement, but should instead be
general enough to permit alternative interconnection arrangements to
evolve.The second concept often used in the context of CDN Federation is
CDN Exchange -- a third-party broker or exchange that is used to
facilitate a CDN federation. Our view is that a CDN exchange offers
valuable machinery to scale the number of CDN operators involved in a
multi-lateral (federated) agreement, but that this machinery is built
on top of the core CDNI mechanisms. For example, as
illustrated in , the exchange might
aggregate and redistribute information about each CDN footprint and
capacity, as well as collect, filter, and redistribute traffic logs
that each participant needs for interconnection settlement, but
inter-CDN Request Routing, inter-CDN content distribution (including
inter-CDN acquisition), and inter-CDN control, which fundamentally
involve a direct interaction between an Upstream CDN and a Downstream
CDN -- operate exactly as in a pair-wise peering arrangement.
Turning to , we observe that in this
example:each CDN supports a direct CDNI Control interface to every
other CDNeach CDN supports a direct CDNI Metadata interface to every
other CDNeach CDN supports a CDNI Logging interface with the CDN
Exchangeeach CDN supports both a CDNI Request Routing interface with
the CDN Exchange (for aggregation and redistribution of dynamic
CDN footprint discovery information) and a direct RI to every
other CDN (for actual request redirection).Note that a CDN exchange may alternatively support a different set
of functionality (e.g., Logging only, or Logging and full request
routing, or all the functionality of a CDN including content
distribution). All these options are expected to be allowed by the
IETF CDNI specifications.There are a number of trust issues that need to be addressed by a
CDNI solution. Many of them are in fact similar or identical to those in
a simple CDN without interconnection. In a standard CDN environment
(without CDNI), the CSP places a degree of trust in a single CDN
operator to perform many functions. The CDN is trusted to deliver
content with appropriate quality of experience for the end user. The CSP
trusts the CDN operator not to corrupt or modify the content. The CSP
often relies on the CDN operator to provide reliable accounting
information regarding the volume of delivered content. The CSP may also
trust the CDN operator to perform actions such as timely invalidation of
content and restriction of access to content based on certain criteria
such as location of the user and time of day, and to enforce per-request
authorization performed by the CSP using techniques such as URI
signing.A CSP also places trust in the CDN not to distribute any information
that is confidential to the CSP (e.g., how popular a given piece of
content is) or confidential to the end user (e.g., which content has
been watched by which user).A CSP does not necessarily have to place complete trust in a CDN. A
CSP will in some cases take steps to protect its content from improper
distribution by a CDN, e.g., by encrypting it and distributing keys in
some out of band way. A CSP also depends on monitoring (possibly by
third parties) and reporting to verify that the CDN has performed
adequately. A CSP may use techniques such as client-based metering to
verify that accounting information provided by the CDN is reliable. HTTP
conditional requests may be used to provide the CSP with some checks on
CDN operation. In other words, while a CSP may trust a CDN to perform
some functions in the short term, the CSP is able, in most cases, to
verify whether these actions have been performed correctly and to take
action (such as moving the content to a different CDN) if the CDN does
not live up to expectations.One of the trust issues raised by CDNI is transitive trust. A CDN
that has a direct relationship with a CSP can now "outsource" the
delivery of content to another (Downstream) CDN. That CDN may in term
outsource delivery to yet another Downstream CDN, and so on.The top-level CDN in such a chain of delegation is responsible for
ensuring that the requirements of the CSP are met. Failure to do so is
presumably just as serious as in the traditional single CDN case. Hence,
an Upstream CDN is essentially trusting a Downstream CDN to perform
functions on its behalf in just the same way as a CSP trusts a single
CDN. Monitoring and reporting can similarly be used to verify that the
Downstream CDN has performed appropriately. However, the introduction of
multiple CDNs in the path between CSP and end user complicates the
picture. For example, third-party monitoring of CDN performance (or
other aspects of operation, such as timely invalidation) might be able
to identify the fact that a problem occurred somewhere in the chain but
not point to the particular CDN at fault.In summary, we assume that an Upstream CDN will invest a certain
amount of trust in a Downstream CDN, but that it will verify that the
Downstream CDN is performing correctly, and take corrective action
(including potentially breaking off its relationship with that CDN) if
behavior is not correct. We do not expect that the trust relationship
between a CSP and its "top level" CDN will differ significantly from
that found today in single CDN situations. However, it does appear that
more sophisticated tools and techniques for monitoring CDN performance
and behavior will be required to enable the identification of the CDN at
fault in a particular delivery chain.We expect that the detailed designs for the specific interfaces for
CDNI will need to take the transitive trust issues into account. For
example, explicit confirmation that some action (such as content
removal) has taken place in a Downstream CDN may help to mitigate some
issues of transitive trust.In general, a CDN has the opportunity to collect detailed information
about the behavior of end users, e.g., by logging which files are being
downloaded. While the concept of interconnected CDNs as described in
this document doesn't necessarily allow any given CDN to gather
more information on any specific user, it potentially facilitates
sharing of this data by a CDN with more parties. As an example, the
purpose of the CDNI Logging interface is to allow a dCDN to share some
of its log records with a uCDN, both for billing purposes as well as for
sharing traffic statistics with the Content Provider on whose behalf the
content was delivered. The fact that the CDNI interfaces provide
mechanisms for sharing such potentially sensitive user data, shows that
it is necessary to include in these interface appropriate privacy and
confidentiality mechanisms. The definition of such mechanisms is dealt
with in the respective CDN interface documents.While there are a variety of security issues introduced by a single
CDN, we are concerned here specifically with the additional issues that
arise when CDNs are interconnected. For example, when a single CDN has
the ability to distribute content on behalf of a CSP, there may be
concerns that such content could be distributed to parties who are not
authorized to receive it, and there are mechanisms to deal with such
concerns. Our focus in this section is on how CDNI
introduces new security issues not found in the single CDN case. For a
more detailed analysis of the security requirements of CDNI, see Section
9 of .Many of the security issues that arise in CDNI are related to the
transitivity of trust (or lack thereof) described in . As noted above, the design of the various interfaces
for CDNI must take account of the additional risks posed by the fact
that a CDN with whom a CSP has no direct relationship is now potentially
distributing content for that CSP. The mechanisms used to mitigate these
risks may be similar to those used in the single CDN case, but their
suitability in this more complex environment must be validated.CDNs today offer a variety of means to control access to content,
such as time-of-day restrictions, geo-blocking, and URI signing. These
mechanisms must continue to function in CDNI environments, and this
consideration is likely to affect the design of certain CDNI interfaces
(e.g., metadata, request routing). For more information on URI signing in
CDNI, see .Just as with a single CDN, each peer CDN must ensure that it is not
used as an "open proxy" to deliver content on behalf of a malicious CSP.
Whereas a single CDN typically addresses this problem by having CSPs
explicitly register content (or origin servers) that are to be served,
simply propagating this information to peer Downstream CDNs may be
problematic because it reveals more information than the Upstream CDN is
willing to specify. (To this end, the content acquisition step in the
earlier examples force the dCDN to retrieve content from the uCDN rather
than go directly to the origin server.)There are several approaches to this problem. One is for the uCDN to
encode a signed token generated from a shared secret in each URL routed
to a dCDN, and for the dCDN to validate the request based on this token.
Another one is to have each Upstream CDN advertise the set of
CDN-Domains they serve, where the Downstream CDN checks each request
against this set before caching and delivering the associated object.
Although straightforward, this approach requires operators to reveal
additional information, which may or may not be an issue.It is noted in that all
CDNI interfaces must be able to operate securely over insecure IP
networks. Since it is expected that the CDNI interfaces will be
implemented using existing application protocols such as HTTP or Extensible Messaging and Presence Protocol (XMPP),
we also expect that the security mechanisms available to those
protocols may be used by the CDNI interfaces. Details of how these
interfaces are secured will be specified in the relevant interface
documents.Digital Rights Management (DRM), also sometimes called
digital restrictions management, is often employed for content
distributed via CDNs. In general, DRM relies on the CDN to distribute
encrypted content, with decryption keys distributed to users by some
other means (e.g., directly from the CSP to the end user). For this
reason, DRM is considered out of scope and
does not introduce additional security issues for CDNI.The following individuals contributed to this document:Matt CaulfieldFrancois Le FaucheurAaron FalkDavid FergusonJohn HartmanBen Niven-JenkinsKent LeungThe authors would like to thank Huw Jones and Jinmei Tatuya for their
helpful input to this document. In addition, the authors would like to
thank Stephen Farrell, Ted Lemon, and Alissa Cooper for their reviews,
which have helped to improve this document.Content Distribution Network Interconnection (CDNI) RequirementsCDNI Request Routing: Footprint and Capabilities SemanticsThis document tries to capture the semantics of the "Footprint and Capabilities Advertisement" part of the CDNI Request Routing interface, i.e. the desired meaning and what "Footprint and Capabilities Advertisement" is expected to offer within CDNI. The discussion in this document has the goal to facilitate the choosing of one or more suitable protocols for "Footprint and Capabilities Advertisement" within CDNI Request Routing.CDNI Control Interface / TriggersThis document describes the part of the CDN Interconnect Control Interface that allows a CDN to trigger activity in an interconnected CDN that is configured to deliver content on its behalf. The upstream CDN can use this mechanism to request that the downstream CDN pre-positions metadata or content, or that it re-validate or purge metadata or content. The upstream CDN can monitor the status of activity that it has triggered in the downstream CDN.CDNI Logging InterfaceThis memo specifies the Logging interface between a downstream CDN (dCDN) and an upstream CDN (uCDN) that are interconnected as per the CDN Interconnection (CDNI) framework. First, it describes a reference model for CDNI logging. Then, it specifies the CDNI Logging File format and the actual protocol for exchange of CDNI Logging Files.CDN Interconnection MetadataThe CDNI Metadata interface enables interconnected CDNs to exchange content distribution metadata in order to enable content acquisition and delivery. The CDNI metadata associated with a piece of content provides a downstream CDN with sufficient information for the downstream CDN to service content requests on behalf of an upstream CDN. This document describes both a base set of CDNI metadata and the protocol for exchanging that metadata.Request Routing Redirection Interface for CDN InterconnectionThe Request Routing Interface comprises of (1) the asynchronous advertisement of footprint and capabilities by a downstream CDN that allows a upstream CDN to decide whether to redirect particular user requests to that downstream CDN; and (2) the synchronous operation of an upstream CDN requesting whether a downstream CDN is prepared to accept a user request and of a downstream CDN responding with how to actually redirect the user request. This document describes an interface for the latter part, i.e. the CDNI request routing/ Redirection Interface.URI Signing for CDN Interconnection (CDNI)This document describes how the concept of URI signing supports the content access control requirements of CDNI and proposes a URI signing scheme. The proposed URI signing method specifies the information needed to be included in the URI and the algorithm used to authorize and to validate access requests for the content referenced by the URI. Some of the information may be accessed by the CDN via configuration or CDNI metadata.