Network Working Group D. Kutscher, Ed.
Internet-Draft NEC
Intended status: Standards Track S. Eum
Expires: August 14, 2013 NICT
K. Pentikousis
Huawei
I. Psaras
UCL
D. Corujo
Universidade de Aveiro
D. Saucez
INRIA
February 10, 2013
ICN Research Challenges
draft-kutscher-icnrg-challenges-00
Abstract
This memo describes research challenges for Information-Centric
Networking. Information-centric networking is an approach to evolve
the Internet infrastructure to directly support this use by
introducing uniquely named data as a core Internet principle. Data
becomes independent from location, application, storage, and means of
transportation, enabling in-network caching and replication.
Challenges include naming, security, routing, system scalability,
mobility management, wireless networking, transport services, in-
network caching, and network management.
Status of this Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on August 14, 2013.
Copyright Notice
Kutscher, et al. Expires August 14, 2013 [Page 1]
Internet-Draft ICN Challenges February 2013
Copyright (c) 2013 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3
2. Problems with Information Distribution Today . . . . . . . . . 4
3. ICN Concepts . . . . . . . . . . . . . . . . . . . . . . . . . 5
4. ICN Research Challenges . . . . . . . . . . . . . . . . . . . 6
4.1. Naming and Security . . . . . . . . . . . . . . . . . . . 6
4.2. Routing and Resolution System Scalability . . . . . . . . 8
4.2.1. Route-By-Name Routing (RBNR) . . . . . . . . . . . . . 9
4.2.2. Lookup-By-Name Routing (LBNR) . . . . . . . . . . . . 9
4.2.3. Hybrid Routing (HR) . . . . . . . . . . . . . . . . . 10
4.3. Mobility Management . . . . . . . . . . . . . . . . . . . 10
4.4. Wireless Networking . . . . . . . . . . . . . . . . . . . 12
4.5. Transport Services . . . . . . . . . . . . . . . . . . . . 12
4.6. In-Network Caching . . . . . . . . . . . . . . . . . . . . 13
4.6.1. Cache Placement . . . . . . . . . . . . . . . . . . . 13
4.6.2. Content Placement -- Content-to-Cache Distribution . . 14
4.6.3. Request-to-Cache Routing . . . . . . . . . . . . . . . 15
4.7. Network Management . . . . . . . . . . . . . . . . . . . . 15
5. Link to and Impact on IETF Technologies . . . . . . . . . . . 17
6. Security Considerations . . . . . . . . . . . . . . . . . . . 17
7. Informative References . . . . . . . . . . . . . . . . . . . . 17
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 19
Kutscher, et al. Expires August 14, 2013 [Page 2]
Internet-Draft ICN Challenges February 2013
1. Introduction
Distributing and manipulating named information is a major
application in the Internet today. In addition to web-based content
distribution, other distribution technologies (such as P2P and CDN)
have emerged and are promoting a communication model of accessing
data by name, regardless of origin server location.
In order to respond to increasing traffic volume in the current
Internet for applications such as mobile video and cloud computing, a
set of disparate technologies and distribution services are applied
that employ caching, replication and content distribution in
different specific ways. These approaches are currently deployed in
separate silos -- different CDN providers and P2P applications rely
on specific distribution technologies. It is not possible to
uniquely and securely identify named information independently of the
distribution channel; and the different distribution approaches are
typically implemented as an overlay, potentially leading to
unnecessary inefficiency.
For example, creating and sharing multimedia content in a social
networking application today, typically requires uploading data
objects to centralized service provider platforms, from where it can
be accessed individually by other users. Even if content sharing is
intended to happen locally, e.g., in a local network or local area,
the actual communication will require interactions from any
interested user with the service provider. CDNs can alleviate the
situation only partly, because, due to organizational and economic
reasons, it is not common to deploy CDN gear ubiquitously. Moreover,
since CDNs and the HTTP communication sessions form overlays, the
actual communication, i.e., the requests for named content and the
actual responses, are largely invisible to the network, i.e., it is
not easily possible to optimize efficiency and performance. For
example in a wireless access network, it is not possible to leverage
inherent broadcast functionality (to avoid duplicate transmission of
the same content) due to limitations from point-to-point and overlay
communication.
Information-centric networking (ICN) is an approach to evolve the
Internet infrastructure to directly support this use by introducing
uniquely named data as a core Internet principle. Data becomes
independent from location, application, storage, and means of
transportation, enabling in-network caching and replication. The
expected benefits are improved efficiency, better support for
provenance verification and name-content binding validation, better
scalability with respect to information/bandwidth demand and better
robustness in challenging communication scenarios.
Kutscher, et al. Expires August 14, 2013 [Page 3]
Internet-Draft ICN Challenges February 2013
ICN concepts can be applied to different layers of the protocol
stack: name-based data access can be implemented on top of the
existing IP infrastructure, e.g., by providing resource naming,
ubiquitous caching and corresponding transport services, or it can be
seen as a packet-level internetworking technology that would cause
fundamental changes to Internet routing and forwarding. In summary,
ICN is expected to evolve the Internet architecture at different
layers.
This document describes research challenges for ICN that need to be
addressed in order to achieve these goals. The objective of this
document is to document these challenges and corresponding current
approaches and to expose requirements that should be addressed by
future research work.
2. Problems with Information Distribution Today
The best current practice to manage this growth in terms of data
volume and devices is to employ application-layer overlays such as
CDNs, P2P applications, and M2M application platforms that cache
content, provide location-independent access to data, and optimize
its delivery. In principle, such platforms provide a service model
of accessing named data objects (NDOs) (replicated web resources, M2M
data in data centers) instead of a host-to-host packet delivery
service model. However, since this functionality resides in overlays
only, the full potential of content distribution and M2M application
platforms cannot be leveraged as the network is not aware of data
requests and data transmissions, leading to:
o data having to travel sub-optimal routes depending on the overlay,
and not the Internet layer, topology;
o multicast and broadcast features of wireless networks cannot be
leveraged, i.e., request and delivery for the same object have to
be made multiple times;
o overlays typically require a significant amount of infrastructure
support, e.g., authentication portals, content storage, and
applications servers, making it often impossible to establish
local, direct communication;
o the network not being aware of the nature of data objects and thus
being unable to manage access and transmission (without layer
violations);
o provenance validation uses host authentication today, so that even
if there are locally cached copies available, it is normally not
Kutscher, et al. Expires August 14, 2013 [Page 4]
Internet-Draft ICN Challenges February 2013
easily possible to validate their authenticity; and
o many applications providing their own approach to caching,
replication, transport, authenticity validation (if at all),
although they all share similar models of accessing named data
objects in the network.
3. ICN Concepts
Fundamentally, ICN is providing access to named data as a first-order
network service, i.e., the network is able to serve requests to named
data natively. That means, network nodes can receive requests for
named data and act on it, for example by forwarding the request to a
suitable next-hop. Consequently, the network processes requests for
named data objects (and corresponding responses) natively, i.e., it
can see requests and responses. Every network nodes on a path is
enabled to perform forwarding decisions, to cache objects etc. This
enables the network to forward such requests on optimal paths,
employing optimal transmission technologies at every node, e.g.,
broadcast/multicast transmission in wireless networks to avoid
duplicate transmission of both requests and responses.
In ICN, like in the Internet Protocol, there is a set of common
concepts and node requirements beyond this basic service model.
Naming data objects is a key concept. In general, ICN names do not
represent neither network nodes nor interfaces -- they represent NDOs
independent of their location. Names are the keys for forwarding
decisions -- and they are used for matching requests to responses: In
order to provide better support for accessing copies of NDOs
regardless of their location, it is important to be able to validate
that a response actually delivers the bits that correspond to an
original request for named data. Name-content binding validation is
a fundamental security service in ICN, and this is often achieved by
establishing a verifiable binding between the object name and the
actual object or an identity that has created the object. ICN can
support other security services, such as provenance validation,
encryption -- depending on the details of naming schemes, object
models and assumptions on infrastructure support. Security services
such as name-content binding validation are available to any node,
i.e., not just the actual receivers. This is an important feature,
for enabling ingress gateways to check object authenticity to prevent
denial-of-service attacks.
Based on these fundamental properties it is possible to leverage
network storage ubiquitously: every node and every device can cache
data objects and respond to requests for such objects -- it is not
required to validate the authenticity of the node itself since name-
Kutscher, et al. Expires August 14, 2013 [Page 5]
Internet-Draft ICN Challenges February 2013
content bindings can be validated. Ubiquitous in-network storage can
be used for different purposes: it can enable sharing, i.e., the same
object copy can be delivered to multiple users/nodes as in today's
proxy caches and CDNs. It can also be used to make communication
more robust (and perform better) by enabling the network to answer
requests from local caches (instead of from origin servers). In case
of disruption (message not delivered), a node can re-send the
request, and it could be answered by an on-path cache, i.e., on the
other side of the disrupted link. The network itself would thus
support retransmissions -- enabling shorter round-trip times and
offloading origin servers and other parts of the network.
The request/response model and ubiquitous in-network storage also
enables new options for implementing transport services, i.e.,
reliable transmission, flow control etc. First of all, a request/
response model can enable receiver-driven transport regimes, i.e.,
receivers (the requestors of NDOs) can control message sending rates
by regulating the request sending rate (assuming that every response
message has to be triggered by a request message). Retransmission
would be achieved by re-sending requests, e.g., after a timeout.
Because objects can be replicated, object transmission and transport
sessions would not necessarily have end-to-end semantics: requests
can be answered by caches, and a node can select one or multiple
next-hop destination for a particular request -- depending on
configuration, observed performance or other criteria.
This receiver-driven communication model potentially enables new
interconnection and business models: a request for named data can be
linked to an interest of a requestor (or requesting network) in data
from another peer, which could suggest modeling peering agreements
and charging accordingly.
4. ICN Research Challenges
4.1. Naming and Security
Naming data objects is as important for ICN as naming hosts is for
today's Internet. Fundamentally, ICN requires unique names for
individual NDOs, since names are used for identifying objects
independently of its location or container. It is important to
establish a verifiable binding between the object and its name (name-
data integrity ), so that a receiver can be sure that received bits
actually represent the NDO (object authenticity). Information about
an object's provenance, i.e., who generated or published it, is also
useful to associate to the name.
The above functions are fundamentally required for the information-
Kutscher, et al. Expires August 14, 2013 [Page 6]
Internet-Draft ICN Challenges February 2013
centric network to work reliably -- otherwise neither network
elements nor receivers can trust objects' authenticity, which would
enable several attacks including critical DoS attacks by injecting
spoofed content into the network. There are different ways to use
names and cryptography to achieve the desired functions [ICNNAMING]
[ICNSURVEY], and there are different ways to manage namespaces
correspondingly.
Two naming schemes have largely been proposed: one with a
hierarchical and one with a flat namespace. The hierarchical scheme
has a structure similar to current URLs, where the hierarchy is
rooted in a publisher prefix. The hierarchy enables aggregation of
routing information, improving scalability of the routing system. In
some cases, the names are human-readable, which makes it possible for
users to manually type in names, reuse, and, to some extent, mapping
name to a user's intent.
The other naming scheme is self-certifying, meaning that the object's
name-data integrity can be verified without needing a public key
infrastructure (PKI) or other third party to first establish trust in
the key. Self-certification is achieved by binding the hash of the
content closely to the object's name. This can be done by directly
embedding the hash of the content in the name. Another option is an
indirect binding, which embeds the public key of the publisher in the
name and signs the hash of the content with the corresponding secret
key. The resulting names are typically non-hierarchical, or flat,
although the publisher field provides structure that can be used for
routing aggregation.
There are design trade-offs for ICN naming affecting routing and
security. Self-certifying names are not human readable nor
hierarchical. They can however provide some structure for
aggregation, for instance, a name part corresponding to a publisher.
Without self-certification, as mentioned above, the infrastructure
depends on a PKI for its operation, which can be impede a large-scale
deployment.
Specific research challenges include:
o naming static data objects can be performed by using content
hashes as part of object names, so that publishers calculate the
hash over existing data objects and receivers (or any ICN node)
can validate the name-content binding by re-calculating the hash
and comparing it to the name (component). [I-D.farrell-decade-ni]
specifies a concrete naming format for this.
o naming dynamic objects is referring to use cases where the name
has to be generated before the object is created (for example,
Kutscher, et al. Expires August 14, 2013 [Page 7]
Internet-Draft ICN Challenges February 2013
this could be the case for live streaming, when a publisher wants
to make the stream available by registering stream chunk names in
the network). One approach to this can be self-certified names as
described above.
o requestor privacy protection can be a challenge in ICN as a direct
consequence of the accessing-named-data-objects paradigm: if the
network can "see" requests and responses, it can also log request
history for network segments or individual users, which can be
undesirable, especially since name are typically expected to be
long-lived. I.e., even if the name itself does not reveal much
information, the assumption is that the name can be used to
retrieve the corresponding data objects in the future.
o Updating and versioning NDO can be challenging because it can
contradict fundamental ICN assumptions: if an NDO can be
replicated and stored in in-network storage for later retrieval,
names have to be long-lived -- and the name-content binding must
not change: updating an object (changing the content without
generating a new name) is impossible. Versioning can be seen as
one possible solution, possibly requiring a naming scheme that
supports versioning (and a way for requestors to learn about
versions).
o Managing accessibility: whereas in ICN the general assumption is
to enable ubiquitous access to NDOs, there can be relevant use
cases where access to objects should be restricted, for example to
a specific user group. There are different approaches for this,
such as object encryption (requiring key distribution and related
mechanisms) or the concept of scopes, e.g., based on names that
can only be used/resolved under some constraints.
4.2. Routing and Resolution System Scalability
ICN routing locates a data object based on its name which is
initially provided by a requester. ICN routing is composed of three
steps: a name resolution step, a discovery step, and a delivery step.
The name resolution step translates the name of requesting data
object into its locator. The discovery step routes user request to
data object based on its name or locator. The last delivery step
routes the data object back to the requester. Depending on how these
steps are combined, ICN routing schemes can be categorized as: Route-
By-Name Routing (RBNR), Lookup-By-Name Routing (LBNR), and Hybrid
Routing (HR).
Kutscher, et al. Expires August 14, 2013 [Page 8]
Internet-Draft ICN Challenges February 2013
4.2.1. Route-By-Name Routing (RBNR)
RBNR omits the first name resolution step. The name of data object
is directly used to route user request to the data object.
Therefore, routing information to each data object basically has to
be maintained in the routing table. Since the number of data objects
is huge (The number of originally published content files that ICN is
expected to support was estimated as 10^11 back in 2007 [DONA].
However, there are still many people in ICN research community who
believe that the number should be larger than 10^11 , e.g. 10^15 --
10^22.), the size of routing table tends to be proportional to the
number of data object unless any aggregation mechanism is introduced
to the name of data object. On the other hand, RBNR reduces overall
latency and simplifies the routing process due to the omission of the
resolution process. For the delivery step, RBNR needs another
identifier (ID) of either host or location to forward the requested
data object back to the requester. Otherwise, an additional routing
mechanism has to be introduced such as bread-crumb routing
[BREADCRUMBS]: a request leaves behind a trail of breadcrumbs along
its forwarding path, and then the response is forwarded back to the
requester consuming the trail. Specific challenges include:
o How to aggregate the names of data objects to reduce the number of
routing entries?
o How does user learn the name which is designed for aggregation by
provider? (For example, although we name our contribution as "ICN
research challenge", IRTF (provider) may want to change the name
to "/IETF/IRTF/ ICN/Research challenge" for aggregation. In this
case, how does a user learn the name "/IETF/IRTF/ICN/Research
challenge" to retrieve the contribution initially named "ICN
research challenge" without any resolution process?)
o Without introducing the name aggregation scheme, can we still
achieve a scalable routing by taking advantage of topological
structure and distributed copies? e.g. compact routing [COMPACT],
random walk [RANDOM] or Greedy routing [GREEDY], etc.
o How to incorporate copies of a data object in in-network caches in
this routing scheme?
4.2.2. Lookup-By-Name Routing (LBNR)
LBNR uses the first name resolution step to translate the name of
requesting data object into its locator. Then, the second discovery
step is carried out based on the locator. Since IP address could be
used as locators, the discovery step can depend on the current IP
infrastructure. The delivery step can be implemented same as IP
Kutscher, et al. Expires August 14, 2013 [Page 9]
Internet-Draft ICN Challenges February 2013
routing. The locator of requester is included in the request
message, and then the requested data object is delivered to the
requester based on the locator. Specific challenges include:
o How to build a scalable resolution system which provides
* Fast lookup: mapping the name of data object to its locators
(copies as well).
* Fast update: the location of data object is expected to change
frequently. Also, multiple data objects may change their
locations at the same time, e.g. data objects in laptop.
o How to incorporate copies of a data object in in-network caches in
this routing scheme?
4.2.3. Hybrid Routing (HR)
HR combines both RBNR and LBNR to benefit from their advantages. For
instance, within a single administrative domain, e.g. ISP where
scalability issue is not serious problem, RBNR can be adopted to
reduce overall latency by omitting the resolution process. On the
other hand, LBNR can be used to route among the domains which have
their own prefix (locator). A specific challenge here is:
o How to design a scalable mapping system, which given the name of
data object, it should return a destination domain locator so that
a user request can be encapsulated and forwarded to the domain?
4.3. Mobility Management
IP was not designed to consider node mobility originally, forcing new
connections towards the content sources to be made. With the
proliferation of mobile terminals equipped with different kinds of
access technologies, mobility became a highly sought solution to be
available at the network layer, berthing Mobile IP [RFC5944] based
protocols. However, this addition also placed a higher degree of
complexity on the network operations due to the need for new network
entities, new signaling messages and resulting side-mechanisms, such
as tunneling. In that sense, novel content-centric network
architectures that go beyond host-based mobility control, provide the
ample grounds for the definition of operating mechanisms considering
mobility as a prime requirement, right at the start.
ICN naming for reaching content intrinsically supports mobility. For
example, CCN [CCN] does not share the IP restriction of forwarding on
spanning trees, so it is able to take advantage of multiple
interfaces or adapt to the changes produced by rapid mobility (i.e.,
Kutscher, et al. Expires August 14, 2013 [Page 10]
Internet-Draft ICN Challenges February 2013
there is no need to bind a layer 3 address into a layer 2 address).
In fact, client mobility is simplified by allowing requests for new
content to normally flow from different interfaces, or through newly
connected points of attachment to the network. However, that
simplicity may not be reflected when the node moving is the content
source, requiring more complex support from the networking mechanisms
in respect to different aspects, such as forwarding update and
caching rebuilding. Furthermore, requirements become more stringent
when support for seamless mobility is required, especially in cases
such as real-time voice/video communications. These requirements are
further exacerbated when mobile nodes are able to connect through
wireless access interfaces of different technologies, where the
performance and link conditions can vary widely depending of numerous
factors.
Here mobility management has an important role in terms of not only
optimizing the handover process, but also to ideally ensure seamless
transition from one point of attachment to the other. In this way,
"seamless transition" ensures that the content reception by the user
occurs in an unperceptive way to the user and/or application
receiving that content. Moreover, this transition needs to be
executed in parallel with ICN content identification and reaching
mechanisms enabling scenarios, such as, preparation of the content
reaching process at the target connectivity point, prior to the
handover (to reduce link switch disturbances). Finally, these
mobility aspects can also be tightly coupled with network management
aspects, in respect to policy enforcement, link control and other
parameters necessary for establishing the node's link to the network.
The resulting mobility management process can thus enhance and evolve
ICN aspects by making them aware (or able to contribute) to not only
allow but also enhance possible mobility procedures.
From this, a set of research challenges on ICN Mobility Management
can be derived:
o How can content reaching mechanisms interface with specific link
operations, such as identifying which links are available for a
certain content
o How to make mobility as a service that is only activated when the
specific user/content/conditions require it (i.e., a possible
solution to maintain the mobility-agnostic aspect of generic ICN)
o How to coordinate mobility management between the node and the
network for optimization and policing procedures?
Kutscher, et al. Expires August 14, 2013 [Page 11]
Internet-Draft ICN Challenges February 2013
4.4. Wireless Networking
Today, all wireless network/radio access technologies (L2) are
developed with a clear assumption in mind: the waist of the protocol
stack is (and will be) IP. This translates into answering a large
set of questions, from how to handle broadcast to how to support
multicast in a rather straightforward manner. Arguably, if one
designs a future wireless access technology with an information-
centric "layer 3", most of these answers would no longer be valid.
Although this is clearly outside the scope of this document, a few
research challenges that the wider community may be interested in
include:
o In the context of wireless access, how can we leverage the
broadcast nature of the medium in an information-centric network?
o Is it possible that by changing the network paradigm to ICN we can
in practice increase the spectral efficiency (bits/s/Hz) of a
wireless network beyond what would be possible with today's host-
centric approaches?
o How can a conversational service be supported at least as
efficiently as today's SoA wireless network deliver?
4.5. Transport Services
ICN's receiver-driven communication model as described above creates
new option for transport protocol design -- it does not rely on end-
to-end communication path from a sender to a receiver, because a
requested object can be accessible in multiple different network
locations. A node can thus decide how to utilize multiple sources,
e.g., by sending parallel requests for the same object or by
switching sources (or next hops) in a suitable schedule for a series
of requests.
In this model the requestor would control data rate by regulating its
request sending rate and next by performing source/next-hop
selections. Specific challenges are depending on the specific ICN
approach in use, but general challenges for receiver-driven transport
protocols (or mechanisms, since dedicated protocols might not be
required) include flow and congestion control, fairness, network
utilization, stability (of data rates under stable conditions) etc.
[HRICP] describes a sample request rate control protocol and
corresponding design challenges.
ICN offers routers the possibility to aggregate requests and can use
several paths, meaning that there is no such thing as end-to-end
communication path, e.g., a router that receives two requests for the
Kutscher, et al. Expires August 14, 2013 [Page 12]
Internet-Draft ICN Challenges February 2013
same content at the same time only sends one requests to its
neighbor. The aggregation of requests has a general impact on
transport service design.
Achieving fairness for requestors can be one challenge as it is not
possible to identify the number of clients behind one particular
request. A second problem related to request aggregation is the
management of request retransmissions. Generally, it is assumed that
a router will not transmit a request if it transmitted an identical
request recently and because there is no information about the
requester, the router cannot distinguish the initial request form a
client from a retransmission from the same client. In such a
situation, how routers can adapt their timers to use the best of the
communication paths. Finally, aggregation of requests has an impact
on the server (producer) side. This last has no way to determine the
number of clients actually consuming the content it is producing.
This shift of model influence the business model of the server, e.g.,
how to implement pay-per-click?
4.6. In-Network Caching
Explicitly named content objects allow for caching in virtually any
network element, including routers, proxy caches and end-host
machines. In-network caching can therefore improve network
performance by fetching content from nodes geographically placed
closer to the end-user. Several issues that need further
investigation have been identified with respect to in-network
caching. Here we list some of the most important challenges that
relate to the properties of the new ubiquitous caching system.
4.6.1. Cache Placement
The declining cost of fast memory gives the opportunity to deploy
caches in network routers and take advantage of explicitly named
cached contents. There exist two approaches to in-network caching,
namely on-path and off-path caching. Both approaches have to
consider the issue of cache location. Off-path caching is similar to
traditional proxy-caching, or CDN server placement. Retrieval of
contents from off-path caches requires redirection of requests and
therefore, is closely related to the Request-to-Cache Routing problem
discussed below. Off-path caches have to be placed in strategic
points within a network in order to reduce the redirection delays and
the number of detour hops to retrieve cached contents. Previous
research on proxy-caching and CDN deployment is helpful in this case.
On the other hand, on-path caching requires less network intervention
and fits more neatly in an information-/content-centric network.
However, on-path caching requires line-speed operation, a fact that
Kutscher, et al. Expires August 14, 2013 [Page 13]
Internet-Draft ICN Challenges February 2013
places more constraints on the design and operation of in-network
caching elements. Furthermore, the gain of such a system of on-path
in-network caches relies on opportunistic/accidental cache hits and
has therefore been considered of limited benefit, given the huge
amount of contents hosted in the Internet. For this reason, network
operators might initially consider only a limited number of network
elements to be upgraded to in-network caching elements. The decision
on which nodes should be equipped with caches is an open issue and
might be based, among others, on topological criteria, or traffic
characteristics. These challenges relate to both the Content
Placement Problem and the Request-to-Cache Routing Problem discussed
next.
4.6.2. Content Placement -- Content-to-Cache Distribution
Given a number of (on-path or off-path) in-network caching elements,
content-to-cache distribution will affect both the dynamics of the
system, in terms of request redirections (mainly in case of off-path
caches) and the gain of the system in terms of cache hits. A
straightforward approach to content placement is on-path placement of
contents as they travel from source to destination. This approach
reduces the computation and communication overhead of placing
contents within the network, but on the other hand might reduce the
chances of hitting cached contents. This relates to the Request-to-
Cache Routing problem discussed next.
Furthermore, the number of replicas held in the system brings up
resource management issues in terms of cache allocation. For
example, continuously replicating content objects in all network
elements results in redundant copies of the same objects. The issue
of redundant replication has been investigated in the past for
hierarchical web-caches. However, in hierarchical web-caching,
overlay systems co-ordination between the data and the control plane
can guarantee increased performance in terms of cache hits. Line-
speed, on-path in-network caching poses different requirements and
therefore, new techniques need to be investigated. In this
direction, there already exist some studies that attempt to reduce
redundancy of cached copies. However, the issue of coordinated
content placement in on-path caches still remains open.
The Content-to-Cache Allocation problem relates also to the
characteristics of the content to be cached. Popular content might
need to be put in places where it is going to be requested next.
Furthermore, issues of "expected content popularity" might need to be
considered in order for some contents to be given priority (e.g.,
popular content vs. one-timers). The criteria as to which contents
should be given priority in in-network content caches relate also to
the business relationships between content providers and network
Kutscher, et al. Expires August 14, 2013 [Page 14]
Internet-Draft ICN Challenges February 2013
operators. Such issues need to be investigated and relate also to
the evaluation methodology discussed later on.
4.6.3. Request-to-Cache Routing
In order to get advantage of cached contents, requests have to be
forwarded to the nodes that temporarily host (cache) the
corresponding contents. This challenge relates to name-based
routing, discussed before. Requests should ideally follow the path
to the cached content. However, instructions as to which content is
cached where cannot be broadcast throughout the network. Therefore,
the knowledge of a content's location at the time of the request
might either not exist, or it might not be accurate (i.e., contents
might have been removed by the time a request is redirected to a
specific node).
Co-ordination between the data and the control planes to update
information of cached contents has been considered, but in this case
scalability issues arise. We therefore, have two options. We either
have to rely on opportunistic caching, where requests are forwarded
to a server and in case the content they are looking for is found on
the path, then content is fetched from this node (instead of the
original server), or we employ cache-aware routing techniques.
Cache-aware routing can either involve both the control and the data
plane, or only one of them. Furthermore, cache-aware routing can be
done in a domain-wide scale or can involve more than one individual
AS. In the latter case, business relationships between ASes might
need to be exploited in order to build a scalable model.
4.7. Network Management
Managing networks has been a core craft in the IP-based host-centric
paradigm ever since the technology was introduced in production
networks. However, at the onset of IP, management was considered
primarily as an add-on. Essential tools that are used daily by
networkers, such as ping and traceroute, did not become widely
available until more than a decade or so after IP was first
introduced. Management protocols, such as SNMP, also became
available much later than the original introduction of IP and many
still consider them insufficient despite the years of experience we
have running host-centric networks. Today, when new networks are
deployed network management is considered a key aspect for any
operator, a major challenge which is directly reflected in higher
OPEX if not done well. If we want ICN to be deployed in
infrastructure networks, development of management tools and
mechanisms must go hand-in with the rest of the architecture design.
Although defining FCAPS for ICN is clearly outside the scope of this
Kutscher, et al. Expires August 14, 2013 [Page 15]
Internet-Draft ICN Challenges February 2013
document, there is a need for creating basic tools early on while ICN
is still in the design and experimentation phases that can evolve
over time and help network operations centers (NOC) to define
policies, validate that they are indeed used in practice, be notified
early on about failures, determine and resolve configuration
problems. AAA as well as performance management, from a NOC
perspective, will also need to be considered. Given the expectations
for a large number of nodes and unprecedented traffic volumes,
automating tasks, or even better employing self-management mechanisms
is preferred. The main challenge here is that all tools we have at
our disposal today are node-centric, end-to-end oriented, or assuming
a packet-stream communication environment. Rethinking reachability
and operational availability, for example, can yield significant
insights into how information-centric networks will be managed in the
future.
With respect to network management we see two different aspects.
First, any operator needs to manage all resources available in the
network, which can range from node connectivity to network bandwidth
availability to in-network storage to multi-access support. In ICN,
users will also bring into the network significant resources in terms
of network coverage extension, storage, and processing capabilities.
DTN characteristics should also be considered to the degree that this
is possible (e.g. content dissemination through data mules). On the
other hand, given that nodes and links are not at the center of an
information-centric network, network management should capitalize on
native ICN mechanisms. For example, in-network storage and name
resolution can be used for monitoring, while native publish/subscribe
functionality can be used for triggering notifications.
However, the considerations on leveraging intrinsic ICN mechanisms
and capabilities to support management operations go beyond a simple
mapping exercise. In fact, not only it raises a series of challenges
on its own, but also opens up new possibilities for both ICN and
"network management" as a concept. For instance, naming mechanisms
are central to ICN intrinsic operations, which are used to identify
and reach content under different aspects (e.g., CCN uses a
hierarchical namespace able to contain human-readable naming scheme,
NetInf uses a flat naming structure, etc.). In this way, ICN is
decoupled from host-centric aspects on which traditional networking
management schemes rely upon. As such, questions are raised which
can directly be translated into challenges for network management
capability, such as, for example how to address a node or a network
segment in a ICN naming paradigm, how to identify which node is
connected "where", and if there is a host-centric protocol running
from which the management process can also leverage upon.
But, on the other hand, these same inherent ICN characteristics also
Kutscher, et al. Expires August 14, 2013 [Page 16]
Internet-Draft ICN Challenges February 2013
allow us to look into network management through a new perspective.
By centering its operations around content, one can conceive new
management operations addressing, for example, per-content management
or access control, as well as analyzing performance per content name
instead of per link or node. Moreover, such considerations can also
be used to manage operational aspects of ICN mechanisms themselves.
For example, [NDN-MGMT] re-utilizes inherent content-centric
capabilities of CCN to manage optimal link connectivity for nodes, in
concert with a network optimization process. Conversely, how these
content-centric aspects can otherwise influence and impact management
in other areas (e.g., security, resilience) is also important, as
exemplified by in [ccn-access], where access control mechanisms are
integrated into a prototype of the [PURSUIT] architecture.
In this way, a set of core research challenges on ICN management can
be derived as:
o Manage and control content reception at the destination
o Coordination of management information exchange and control
between ICN nodes and ICN network control points Identification of
management and controlling actions and items through information
naming
o Relationship between NDOs and host entities identification (i.e.,
how to identify a particular link, interface or flow that need to
be managed)
5. Link to and Impact on IETF Technologies
TBW later.
6. Security Considerations
See naming and security challenges.
7. Informative References
[BREADCRUMBS]
Rosensweig, E. and J. Kurose, "Breadcrumbs: Efficient,
Best-Effort Content Location in Cache Networks",
In Proceedings of the IEEE INFOCOM 2009, April 2009.
[CCN] Jacobsen, K, D, F, H, and L, "Networking Named Content",
CoNEXT 2009 , December 2009.
Kutscher, et al. Expires August 14, 2013 [Page 17]
Internet-Draft ICN Challenges February 2013
[COMPACT] Cowen, L., "Compact routing with minimum stretch",
In Journal of Algorithms, vol. 38, pp. 170--183, 2001.
[DONA] Koponen, T., Ermolinskiy, A., Chawla, M., Kim, K., gon
Chun, B., and S. Shenker, "A Data-Oriented (and Beyond)
Network Architecture", In Proceedings of SIGCOMM 2007,
August 2007.
[GREEDY] Papadopoulos, F., Krioukov, D., Boguna, M., and A. Vahdat,
"Greedy forwarding in dynamic scale-free networks embedded
in hyperbolic metric spaces", In Proceedings of the IEEE
INFOCOM, San Diego, USA, 2010.
[HRICP] Carofiglio, G., Gallo, M., and L. Muscariello, "Joint hop-
by-hop and receiver-driven interest control protocol for
content-centric networks", In Proceedings of ACM SIGCOMM
ICN 2012, DOI 10.1145/2342488.2342497, 2012.
[I-D.farrell-decade-ni]
Farrell, S., Kutscher, D., Dannewitz, C., Ohlman, B.,
Keraenen, A., and P. Hallam-Baker, "Naming Things with
Hashes", draft-farrell-decade-ni-10 (work in progress),
August 2012.
[ICNNAMING]
Ghodsi, A., Koponen, T., Rajahalme, J., Sarolahti, P., and
S. Shenker, "Naming in Content-Oriented Architectures",
In Proceedings ACM SIGCOMM Workshop on Information-Centric
Networking (ICN), 2011.
[ICNSURVEY]
Ahlgren, B., Dannewitz, C., Imbrenda, C., Kutscher, D.,
and B. Ohlman, "A Survey of Information-Centric
Networking", In Communications Magazine, IEEE , vol.50,
no.7, pp.26-36, DOI 10.1109/MCOM.2012.6231276, 2012.
[NDN-MGMT]
Corujo, D., Aguiar, R., Vidal, I., and J. Garcia-Reinoso,
"A named data networking flexible framework for management
communications", Communications Magazine, IEEE , vol.50,
no.12, pp.36-43 , December 2012.
[PURSUIT] Fotiou et al., N., "Developing Information Networking
Further: From PSIRP to PURSUIT", In Proceedings of Proc.
BROADNETS. ICST, 2010.
[RANDOM] Gkantsidis, C., Mihail, M., and A. Saberi, "Random walks
in peer-to-peer networks: algorithms and evaluation",
Kutscher, et al. Expires August 14, 2013 [Page 18]
Internet-Draft ICN Challenges February 2013
In Perform. Eval., vol. 63, pp. 241--263, 2006.
[RFC5944] Perkins, C., "IP Mobility Support for IPv4, Revised",
RFC 5944, November 2010.
[ccn-access]
Fotiou, N., Marias, G., and G. Polyzos, "Access control
enforcement delegation for information-centric networking
architectures", In Proceedings of the second edition of
the ICN workshop on Information-centric networking (ICN
'12). ACM, New York, NY, USA, 85-90., 2012.
Authors' Addresses
Dirk Kutscher (editor)
NEC
Kurfuersten-Anlage 36
Heidelberg,
Germany
Phone:
Email: kutscher@neclab.eu
Suyong Eum
National Institute of Information and Communications Technology
4-2-1, Nukui Kitamachi, Koganei
Tokyo 184-8795
Japan
Phone: +81-42-327-6582
Email: suyong@nict.go.jp
Kostas Pentikousis
Huawei Technologies
Carnotstrasse 4
Berlin 10587
Germany
Email: k.pentikousis@huawei.com
Kutscher, et al. Expires August 14, 2013 [Page 19]
Internet-Draft ICN Challenges February 2013
Ioannis Psaras
University College London, Dept. of E.E. Eng.
Torrington Place
London WC1E 7JE
United Kingdom
Email: i.psaras@ucl.ac.uk
Daniel Corujo
Universidade de Aveiro
Instituto de Telecomunicacoes, Campus Universitario de Santiago
Aveiro P-3810-193
Portugal
Email: dcorujo@av.it.pt
Damien Saucez
INRIA
Email: damien.saucez@inria.fr
Kutscher, et al. Expires August 14, 2013 [Page 20]