Internet Engineering Task Force H. Xie
Internet-Draft Huawei & USTC
Intended status: Informational T. Tsou
Expires: December 29, 2012 Huawei (USA)
D. Lopez
Telefonica I+D
H. Yin
Huawei (USA)
June 27, 2012
Use Cases for ALTO with Software Defined Networks
draft-xie-alto-sdn-use-cases-01.txt
Abstract
The introduction of SDN fundamentally changes the way that the ALTO
works. This draft describes the Vertical Architecture and the
Horizontal Architecture allowing coherent coexistence of application
layer traffic optimization (ALTO) with software defined network
(SDN). Unique requirements for design and operations are identified
and summarized, suggesting that the Vertical Architecture allows
better division, management, flexibility, privacy control and long-
term evolution of the network. We also define the main interactions
and information flows, and present a set of use cases to illustrate
how we extend ALTO to support SDN, in the Vertical Architecture.
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 December 29, 2012.
Copyright Notice
Copyright (c) 2012 IETF Trust and the persons identified as the
document authors. All rights reserved.
Xie, et al. Expires December 29, 2012 [Page 1]
Internet-Draft ALTO Use Cases For SDN June 2012
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. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Xie, et al. Expires December 29, 2012 [Page 2]
Internet-Draft ALTO Use Cases For SDN June 2012
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1. Terminology . . . . . . . . . . . . . . . . . . . . . . . 5
2. An Overview of Software Defined Network . . . . . . . . . . . 5
2.1. Software Defined Network . . . . . . . . . . . . . . . . . 5
2.2. Division of Network . . . . . . . . . . . . . . . . . . . 6
2.3. SDN Domain . . . . . . . . . . . . . . . . . . . . . . . . 8
3. Architectural Considerations for SDN and ALTO . . . . . . . . 9
3.1. The Vertical Architecture . . . . . . . . . . . . . . . . 9
3.2. The Horizontal Architecture . . . . . . . . . . . . . . . 11
3.3. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 11
4. Interactions between SDN and ALTO . . . . . . . . . . . . . . 12
4.1. ALTO Scopes . . . . . . . . . . . . . . . . . . . . . . . 12
4.2. ALTO clients . . . . . . . . . . . . . . . . . . . . . . . 13
4.3. Interactions between SDN and ALTO . . . . . . . . . . . . 14
4.3.1. Downward Interaction . . . . . . . . . . . . . . . . . 14
4.3.2. Upward Interaction . . . . . . . . . . . . . . . . . . 14
4.4. Interactions between Legacy ALTO Clients and ALTO
Servers . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.5. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 15
5. Information Flows . . . . . . . . . . . . . . . . . . . . . . 15
5.1. Information Flow of SDN Controller . . . . . . . . . . . . 16
5.2. Information Flow of Applications, SDN and ALTO . . . . . . 16
5.2.1. SDN-aware Applications . . . . . . . . . . . . . . . . 16
5.2.2. SDN-unaware Applications . . . . . . . . . . . . . . . 17
5.2.3. Legacy Applications . . . . . . . . . . . . . . . . . 17
5.3. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 18
6. Use Case for Upward Flow . . . . . . . . . . . . . . . . . . . 18
6.1. Unrestrictive Information Exporting . . . . . . . . . . . 18
6.2. Restrictive Information Exporting . . . . . . . . . . . . 19
6.3. Information Aggregation . . . . . . . . . . . . . . . . . 19
7. Use Case for Downward Flow . . . . . . . . . . . . . . . . . . 20
7.1. SDN-Aware QoS Metrics . . . . . . . . . . . . . . . . . . 20
7.1.1. Use Case: On-Demand Bandwidth . . . . . . . . . . . . 21
7.1.2. Delay . . . . . . . . . . . . . . . . . . . . . . . . 22
7.2. Content Delivery Networks (CDN) . . . . . . . . . . . . . 22
7.3. Information-Centric Content Delivery Networks (IC-CDN) . . 24
8. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 26
9. Contributors . . . . . . . . . . . . . . . . . . . . . . . . . 26
10. Acknowlegements . . . . . . . . . . . . . . . . . . . . . . . 26
11. Security Considerations . . . . . . . . . . . . . . . . . . . 26
12. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 27
13. Informative References . . . . . . . . . . . . . . . . . . . . 27
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 27
Xie, et al. Expires December 29, 2012 [Page 3]
Internet-Draft ALTO Use Cases For SDN June 2012
1. Introduction
The concept of Software Defined Network (SDN) has emerged and become
one of the fundamental, promising networking primitives that allow
flexibility of control, separation of functional planes and
continuous evolution in networks.
One of the key features of SDN is the full separation of two
functional planes: the control plane and the data-forwarding plane.
SDN requires that networking devices support such separation,
implementing the data plane mechanisms, while the control plane is
provided by an external entity called the "controller". The other
key feature of SDN is the new interaction process between the
separated control plane and data-forwarding plane. This interaction
is mandated to be performed specific open protocols, allowing for a
free combination of networking devices and controllers, as well as
supporting the controller to take decisions on information additional
to the networking device status.
There could be numerous benefits as a result of the above features in
SDN, e.g., just to name a few, network virtualization, better
flexible control and utilization of the network, networks customized
for scenarios with specific requirements. For instance, some SDN
technologies have started to find their ways into Data Center
Networks (DCNs). Furthermore, in order to allow efficient and
flexible implementation of the above separation and interactions, a
significant portion of the SDN system could typically be implemented
in software, as opposed to the hardware-based implementation adopted
by most of today's networking devices.
Due to the great potentials of SDN and the unique requirements of
DCNs, Data Centers are likely to become a first place where SDN could
be deployed. We envision that SDN could be gradually adopted by
enterprise networks and then by carrier networks due to its unique,
attractive features. When deploying SDN in large-scale distributed
networks, we expect to see a collection of deployments limited in
relatively small segments of a bigger network. In other words, it is
likely that the operator of a large-scale enterprise / carrier
network prefers to divide the whole networks into multiple smaller
segments and put each of there segments in an SDN domain. These
smaller network segments (therefore their corresponding SDN domains)
are connected and jointly form the larger whole network. Such a
divide-and-conquer methodology not only allows gradual deployment and
continuous evolution, but also enables flexible provisioning of the
network.
With the deployment of SDN, application layer traffic optimization
(ALTO) faces new challenges including, but not limited to, privacy
Xie, et al. Expires December 29, 2012 [Page 4]
Internet-Draft ALTO Use Cases For SDN June 2012
preservation, granularity of information collection and exchange,
join optimization, etc. We shall elaborate these challenges and
their impacts on the design of ALTO extensions for SDN in this draft.
1.1. Terminology
While the definition of software defined networks is still "nebulous"
to some extent, we refer to as SDN the networks that implement the
separation of the control and data-forwarding planes and software
defined interactions between these two separated planes; such
interactions are characterized by open interfaces which allow
programming the network equipment's forwarding plane by external
agents, e.g., SDN controllers. However, how the two separated planes
interact is not a focus of this draft; instead, the ALTO extension
for SDN recommended in this draft is independent of how such
interactions would be.
An SDN domain is a portion of a network infrastructure, consisting of
numerous connected networking devices that are SDN capable (i.e., SDN
capable devices implement the control/forwarding plane separation and
the open interfaces allowing external agents to program the devices)
and a domain controller which implements the SDN control plane
functionalities for these devices. An SDN domain can be as small as
a sub-network of a dozen devices or as large as a medium/large data
center network.
A software defined network is a network that has one or multiple SDN
domains. Due to an SDN domain typically has limited coverage, an SDN
may be comprised of networking devices under control of some SDN
domains (i.e., SDN managed devices) and devices under no control of
any SDN domain (i.e., normal devices). Note that such normal devices
could still be SDN capable and their control/forwarding planes are
managed as in normal networks today. This implies that a network
with both normal devices and SDN capable devices (managed by SDN
domains) needs both normal and SDN capable control/forwarding plane
management.
2. An Overview of Software Defined Network
This section provides a high level and conceptual overview of
software defined network in order to help illustrate its unique
requirements for ALTO.
2.1. Software Defined Network
We refer to as an "SDN" a carrier's or an enterprise's network which
deploys or implements software defined networking technologies.
Xie, et al. Expires December 29, 2012 [Page 5]
Internet-Draft ALTO Use Cases For SDN June 2012
Since SDN separates the control plane and data-forwarding plane, we
refer to the entity that implements the control-plane functionalities
as the "SDN controller".
In order for a network to be classified as an SDN, it is unnecessary
that all networking devices have to be SDN capable. Some of devices
in a network may be managed by an SDN controller while the remaining
ones may not; such a network is still qualified as an SDN.
Applications in software defined networks are either SDN-aware or
unaware of SDN.
o If an application is SDN-aware, then the application may prefer
direct communication with SDN controllers, which implies that
there must exist mechanism(s) for SDN-aware applications to locate
and communication with SDN controllers. If the application
prefers indirect communication with SDN controllers, then it
follows the case of SDN-unaware applications (see below).
Applications that are SDN-aware may be able to better utilize the
SDN capable network due to its knowledge about SDN and its
capability of proactive, direct interaction with SDN.
o If an application is SDN-unaware, then the application indirectly
communicates with SDN controllers by sending application datagrams
with specific formats, via which the application can specify its
requirements for the network resources. Legacy applications
(applications for the current IP networks) are largely SDN-
unaware, and such applications may not be able to utilize the SDN
capable networks as efficiently as SDN-aware applications.
Whether and how applications should/can be SDN-aware or SDN-unaware
is beyond the scope of this draft. However, the framework we
described in this draft is applicable to both SDN-aware and SDN-
unaware cases.
2.2. Division of Network
A network operator may decide to divide the network into multiple
sub-networks, some of which are SDN capable and thus are managed by
corresponding SDN controllers.
There could be numerous reasons for such division of network. Below
we summarize a few of them:
o Scalability.
The number of devices an SDN controller can feasibly manage is
likely to be limited. Therefore, in order to manage a many
Xie, et al. Expires December 29, 2012 [Page 6]
Internet-Draft ALTO Use Cases For SDN June 2012
devices that cannot be put under control of a single SDN
controller, multiple controllers have to be deployed. Hence, the
network is divided into multiple sub-networks; if a sub-network
has SDN capable devices, it should be managed by an SDN
controller.
o Manageability.
At the network level, in order to reduce the complexity of
management, a carrier may decide to divide the network into
multiple sub-networks and put some of them under control of some
SDN controllers (provided that the devices in such sub-networks
are SDN capable); each of the sub-networks can be managed
independently to some extent, thus reducing the overall complexity
of managing the whole network. Even at the sub-network level, a
carrier may still decide to further divide the sub-network in
order to further reduce complexity of management. For instance, a
sub-network under control of an SDN controller may span across a
large geographical area or cover a large number of devices; in
this case it is reasonable for the carrier to further divide it
into two or even more sub-networks, such that the complexity of
managing each individual sub-network plus the overall complexity
of managing all divided sub-networks are reduced.
o Privacy.
When a carrier divides its network into multiple sub-networks and
put them under control of SDN, the carrier may choose to implement
difference privacy policies in different sub-networks. For
example, a carrier may dedicate a part of its infrastructure to
some certain customers, dividing the whole network and dedicate
some of the sub-networks is a convenient and scalable way to
manage the network resources for these customers. Another example
is that a sub-network in a carrier's network may be dedicated to
some certain customers and such information as network topology
may not be disclosed to any external entity.
o Deployment
The deployment of network infrastructures, especially new network
infrastructure and technologies, has to be incremental. This
means that at any time, a carrier's network may consist of a
portion of legacy and a portion of non-legacy infrastructure.
When deploying new infrastructure or technologies, it is highly
preferable to limit the scope of new deployment and do it in an
incremental way. In such cases, it is very favorable to divide
the network into multiple individually manageable sub-networks and
choose some of them to deploy the new infrastructure /
Xie, et al. Expires December 29, 2012 [Page 7]
Internet-Draft ALTO Use Cases For SDN June 2012
technologies.
2.3. SDN Domain
With the introduction of SDN, we believe that it is inevitable for
carriers to divide their networks for many reasons as illustrated in
the preceding subsection. Therefore, we believe that to better suit
this need, it should be recommended that SDN domains are defined and
applied in division of the networks.
An SDN domain is a sub-network, resulted from division of a network
which is determined by the network operator. Each domain typically
consists of numerous connected networking devices, each of which is
SDN capable. Each domain also has a domain controller which
implements SDN control plane functionalities for the devices in this
domain. Another important task such a domain controller is
responsible for includes fine-grain network information collection
(for devices in this domain). The collected information is necessary
for the controller to make data-forwarding plane decisions. Note
that such a domain controller may be integrated as a part of a so-
called "network operating system" (NOS). An example of such a
network operating system is illustrated in [1].
Any networking device, if under the control of certain SDN domains,
should belong to only one SDN domain and should be under the control
of the SDN domain's controller. In other words, the intersection of
two domains is always empty.
Furthermore, SDN domains are connected (via the connectivity among
underlying devices provided by the underlying network; such devices
belong to different SDN domains) to form the whole network. For a
large-scale distributed networks owned by a national/multi-national
carrier or enterprise, it is natural to adopt the "divide-and-
conquer" strategy and divide the whole network into multiple SDN
domains. Even for small or medium networks, multiple SDN domains may
be necessary in order to virtualize the network resources (e.g., set
up multiple SDN domains for a large data center network to
instantiate multiple virtual data centers for many content
providers). Note that how multiple SDN domains inside a carrier's/
enterprise' network work together is beyond the scope of this draft
and is handled by other working groups.
Inside each SDN domain, its controller may define domain-specific
policies on information importing from devices, aggregating, and
exporting to external entities. Such policies may not be made
public; therefore, other domain controllers or ALTO may not know the
existence of such policies for any given SDN domain.
Xie, et al. Expires December 29, 2012 [Page 8]
Internet-Draft ALTO Use Cases For SDN June 2012
The natural network division implemented by SDN domains impose
significantly new and challenging requirement for shaping the
interactions between SDN and ALTO, and therefore designing the
protocols to enable such interactions.
3. Architectural Considerations for SDN and ALTO
We introduce two architectures that allow coherent co-existence of
SDN and ALTO in this section:
o the Vertical Architecture (or the V Architecture for short) allows
better division, management, flexibility, privacy control and
long-term evolution of the network, and
o the Horizontal Architecture (or the H Architecture for short)
simplifies the implementation of ALTO extensions for SDN.
We next present each of these two architectures individually.
3.1. The Vertical Architecture
The Vertical Architecture is illustrated in the following figure.
The network has one or multiple SDN domains and an ALTO server. The
interactions between the SDN controllers and the SDN capable devices
fall in the scope of SDN and therefore is out of the scope of this
draft; instead, the interactions between the SDN domains (more
specifically, SDN controllers) and the ALTO server is what this draft
focuses on.
.----------------------------------------------.
| ALTO Server |
`----------------------------------------------'
^ |
.-------------|------------|-------------------.
| | V |
| .-------------------------------. |
| | SDN Controller | |
| `-------------------------------' |
| | |
| | |
| .-------------------------------. |
| | SDN Capable Devices | |
| `-------------------------------' |
| |
| An SDN Domain |
`----------------------------------------------'
Xie, et al. Expires December 29, 2012 [Page 9]
Internet-Draft ALTO Use Cases For SDN June 2012
The Vertical Architecture is a hierarchical architecture consisting
of three tiers. In the first tier, the only entity is the ALTO
server. In the second tier, the only entities are the SDN domain
controllers. In the third tier, the only entities are SDN domains
(each domain consists of numerous networking devices). In this
architecture, all entities are owned by the same carrier. However,
some of the SDN domains (and the networking devices in them) may be
dedicated to certain customers of the carrier (the carrier gives the
customers privileges access). This is subject to a collaboration
agreement between them.
The interactions between the SDN controllers and the ALTO server can
be divided into two categories:
o Upward interactions (i.e., from SDN controllers to ALTO servers):
each SDN controller collects network information from the devices
managed by it and information from other SDN controllers), and
report such information to the ALTO server, subject to the
information aggregation and privacy policies defined for the
corresponding individual SDN domain. Such network information is
referred to the inter-domain network information. The network
information could include key information such as domain-level
network cost, bandwidth, domain-specific connectivity, etc. The
upward interactions could be implemented in either the push model
or the pull model. It is worth noting that network information
collection has not been explored, and that network information
collection could introduce significant overhead and complexity, in
the current scope of ALTO. However, automated network information
collection is a key to the success of ALTO. With the help of SDN
and the Vertical Architecture, such automated network information
collection becomes feasible and appealing. Note that this does
not exclude the possibility of network operators manually or
automatically update the ALTO server with the network information
(e.g., the network cost map). It is also worth noting that an SDN
controller may choose to report its domain-specific network
information only (referred to as the intra-domain information),
with or without privacy policies. In this case, SDN controllers
become an automated information collector for the ALTO server.
o Downward interactions (i.e., from ALTO servers to SDN
controllers): each SDN controller is also an ALTO client and
retrieves relevant network information from the ALTO server. This
is similar to the current scope of ALTO without the existence of
SDN; however, the differences are that with the existence of SDN,
the network information is generally specific to SDN and SDN
domains; SDN controllers as ALTO clients could query the ALTO
server for either inter-domain or intra-domain network information
(provided that intra-domain information is reported and made
Xie, et al. Expires December 29, 2012 [Page 10]
Internet-Draft ALTO Use Cases For SDN June 2012
available).
We refer to as the "upward flow" the information flow from the second
tier (SDN controllers as ALTO clients) to the first tier (ALTO
server), and refer to as the "downward flow" the information flow in
the reverse direction, i.e., from the first tier (ALTO server) to the
second tier (SDN controllers as ALTO clients).
3.2. The Horizontal Architecture
The Horizontal Architecture is illustrated in the following figure.
Each SDN controller is integrated with an ALTO server. The
interactions between the SDN controllers and the ALTO server is
represented by the horizontal line in the figure. An example of this
architecture can be found in [2].
.---------------------------------------------------------------------.
| .--------------------------. .--------------------------. |
| | SDN Controller |<----| ALTO Server | |
| `--------------------------' `--------------------------' |
`-----------------|---------------------------------------------------'
|
.-------------------------------.
| SDN Capable Devices |
`-------------------------------'
In the Horizontal Architecture, the SDN controller can act as an ALTO
client and query the network information of the networking devices
from the ALTO server. However, such network information may not meet
the SDN controllers' granularity requirement (i.e., the information
provided by the ALTO server may not be as fine-grained as needed by
the SDN controllers). In addition, there may exist redundant
information collection, as SDN controllers are inevitably collecting
various fine-grain information from the devices they manage; the
information collection by the ALTO server, either mannully or
automatically, is likely to be redundant. Furthermore, when the
carrier decides to divide its network into multiple SDN domains, it
can be difficult, if not impossible at all, for the Horizontal
Architecture to adapt to the network division.
3.3. Summary
The ALTO server covers a carrier's network as a whole; however, if
the carrier divide the network into multiple SDN domains, each SDN
domain covers only a segment of the network. Additionally, the ALTO
server has only relatively coarse-grained information, while SDN
domain controllers could easily collect more fine-grained
Xie, et al. Expires December 29, 2012 [Page 11]
Internet-Draft ALTO Use Cases For SDN June 2012
information. More importantly, different SDN domains may implement
different information aggregation and privacy policies (e.g., when
such domains are dedicated to certain customers of the carrier).
These observations suggest that the Vertical Architecture is
advantageous over the Horizontal Architecture. With the Vertical
Architecture, SDN and ALTO are explicitly separated and as a result
the logic is cleaner and this allows them to evolve independently.
Furthermore, the Vertical Architecture makes automated information
collect possible for ALTO, which could make ALTO deployment and
management easier and more attractive. Therefore, in the remainder
of this draft, we mainly focus on the Vertical Architecture. We will
describe the interactions and the information flows in further
details for the Vertical Architecture.
4. Interactions between SDN and ALTO
We now describe the interactions between SDN and ALTO in details. We
first compare the ALTO scopes without and with the existence of SDN,
and then describe the various interactions existing in the Vertical
Architecture.
4.1. ALTO Scopes
The existence of SDN differentiates two scopes of ALTO, namely, the
current scope of ALTO without SDN (referred to as the SDN-unfriendly
Scope) and the new scope of ALTO with coherent coexistence of SDN
(referred to as the SDN-friendly Scope):
o the SDN-unfriendly Scope: in the current scope of ALTO, there
exist interactions between ALTO clients and ALTO servers. Such
interactions are one-way interaction, meaning that ALTO
information flows in one direction, i.e., from the server to the
clients. More specifically, ALTO clients submit ALTO requests to
(and subsequently receive ALTO responses from) an ALTO server.
Additionally, ALTO clients in the current scope of ALTO are
network applications who would like to consume the network
resources. ALTO clients are typically managed by network
applications rather than by network carriers. However, ALTO
servers are owned and managed by network carriers.
o With the introduction of SDN, the interactions between ALTO
clients and ALTO servers become more complex. A more careful
examination as well as important ALTO extensions are necessary to
make ALTO work in the context of SDN. It is important to note
that if the network does not implement network division (i.e.,
does not implement SDN domains), the interactions are "compressed"
Xie, et al. Expires December 29, 2012 [Page 12]
Internet-Draft ALTO Use Cases For SDN June 2012
into a compact set of interactions; specifically, both the SDN
controller (recall that there exists only one single controller
for the whole network) and the ALTO server could be integrated in
many equally efficient fashions. For instance, ALTO server could
be put underneath the controller, i.e., ALTO server provides
information to the controller, as suggested by [2]. However, if
the network implements division via SDN domains (i.e., there
exists multiple SDN domains), we essentially "unfold" the
compressed interaction space and need more complex structures that
allow efficient design and implementation, due to the facts that
we listed in the preceding sections. Furthermore, the design
originally for the compressed space could be instantiated for the
unfolded space but could not be as efficient.
We next focus on the SDN-friendly Scope and highlight the complex
structures and the important differences.
4.2. ALTO clients
With the existence of SDN and SDN controllers, ALTO clients could be
not only legacy network applications (or proxies for these network
applications), but also SDN controllers.
In SDN, SDN controllers have similar needs as the legacy ALTO clients
which communicate with ALTO servers, because ALTO clients would like
to better understand the network and thus improve the application
performance.
There are multiple reasons for this similarity. The most important
reason is that each SDN controller is only responsible for managing a
sub-network in a carrier's network; therefore, it may not understand
well other sub-networks in the same carrier network. However, in
order to allocate the network resources to satisfy application
requirements (note that the end points of such applications may well
span across multiple SDN domains), an SDN controller may have to
involve other SDN controllers because the network paths needed may
traverse multiple SDN domains. Thus, obtaining global information
from the ALTO server is a significantly more efficient and more
preferable than otherwise from SDN interconnection protocols; plus,
such protocols do not exist yet today.
Although SDN controllers have similar needs as legacy ALTO
applications do, the fundamental properties of SDN controllers as
ALTO clients are significantly different. One of the differences is
the ownership and management, as most SDN controllers require
additional (and more likely complex) access privileges.
Specifically, SDN controllers are typically owned by the network
carriers who also own their ALTO servers, while legacy ALTO clients
Xie, et al. Expires December 29, 2012 [Page 13]
Internet-Draft ALTO Use Cases For SDN June 2012
are network applications typically not owned by network carriers
(there are cases where owned collaboratively amongst operators).
In terms of management, the entity managing SDN controller is the
same as that managing the ALTO server. We emphasize that when an SDN
domain is dedicated to some customers of a network carrier, the use
of the SDN domain is owned by these customers and so is the
management. In this case, the SDN controllers as ALTO clients are
slightly more like legacy ALTO clients. However, there still exist
fundamental differences which we will illustrate later.
4.3. Interactions between SDN and ALTO
Another difference is the interactions between SDN controllers as
ALTO clients and ALTO servers. Legacy ALTO clients retrieve
information from ALTO servers. However, SDN controllers may also
need to push information to ALTO servers. In a carrier's network,
SDN controllers and the ALTO server are owned by the same carrier.
Interactions between them could be significantly more complex.
4.3.1. Downward Interaction
The downward interaction is the information flow from ALTO servers to
ALTO clients (i.e., SDN controllers). SDN controllers request
information from ALTO servers, similar to legacy ALTO clients.
However, the requested information is significantly different. The
fundamental difference is a result of SDN and SDN domain division,
which do not exist in legacy network application scenarios. For
instance, an SDN controller for a specific SDN domain may be
interested in obtaining internal information of other SDN domains
(provided that other domains allow to do so), or obtaining domain-
level information such as inter-SDN-domain connectivity. None of
these is applicable to legacy ALTO client scenarios. As a result,
ALTO server and its protocol should be extended to support such
scenarios.
4.3.2. Upward Interaction
The upward interaction is the information flow from ALTO clients
(i.e., SDN controllers) to ALTO servers. SDN controllers open up the
possibilities of conveniently collecting network information and
exporting such information to ALTO servers. SDN controllers are at
the best position to collect network information. More importantly,
it is an evitable requirement that SDN controllers collect the
information of the networking devices in its domain.
In some scenarios, it is a requirement that information flow from
ALTO clients to ALTO servers. For instance, an SDN domain could be
Xie, et al. Expires December 29, 2012 [Page 14]
Internet-Draft ALTO Use Cases For SDN June 2012
dedicated to some of a carrier's certain customers; the usage of such
a domain gives privileged client access. However, such a domain is
an integral sub-network of the carrier's network. In such a case,
the ALTO server for the carrier's network is not able to collect
necessary information in a scalable, manageable way. Even if we
assume that the ALTO server can automatically pull necessary
information directly from networking devices, the dedicated domain
may disallow the ALTO server to do so, because customers who own and
manage this domain may enforce stringent privacy policies and
disallow exporting information externally. The SDN controller is the
best entity that can facility the automation of information
collection while at the same time enforce the specific privacy
policy.
4.4. Interactions between Legacy ALTO Clients and ALTO Servers
With the existence of SDN, the way that legacy network applications
(i.e., as legacy ALTO clients) interact with ALTO servers is also
different.
In legacy ALTO client/server scenarios, ALTO clients obtain cost maps
from ALTO servers, with the implicit assumption that ALTO servers
understand how the underlying network routes packets, which allows
ALTO servers to define or compute a cost metric associated with a
given route.
However, with the introduction of SDN, such assumption may no longer
hold, as SDN controllers may dynamically negotiate and determine a
route between two end points (which may belong to two different SDN
domains), especially when applications have specific requirements for
network resources (e.g.bandwidth, delay, etc). Thus, in order for
applications to best utilize the network resources, the way that
legacy ALTO clients communicate with ALTO servers should be adapted
to SDN.
4.5. Summary
In the context of SDN, due to the specific and unique properties of
SDN domains, SDN controllers as ALTO clients are significantly
different from legacy ALTO clients, posing new requirements for the
interactions between ALTO clients and ALTO servers.
5. Information Flows
We now further describe the two different information flows through
two sets of use cases, one for the information flow from ALTO servers
to ALTO clients, the other for the information flow from SDN
Xie, et al. Expires December 29, 2012 [Page 15]
Internet-Draft ALTO Use Cases For SDN June 2012
controllers to ALTO servers.
5.1. Information Flow of SDN Controller
A network may consist of multiple SDN domains. Note that due to
operational or deployment concerns, there may exist networking
devices that do not belong to any SDN domain. In each SDN domain,
the SDN controller is responsible for the following tasks (only ALTO
related tasks are included below):
o Collect fine-grain information from the networking devices it
manages. Such information could include, but not limited to, SDN
domain topology, link capacity, available bandwidth on each link,
links connected to external devices belonging to other SDN
domains.
o Implement pre-defined domain-specific policies. Such policies
could include, but not limited to, how resources should be
allocated, how the collected information should combined and
presented.
o Optionally aggregate the collected information for external view
purposes per its policies.
o Obtain cost maps at the granularity of SDN domains or obtain
internal cost maps for specific domains (if available), consult
for cross-domain data-forwarding plane recommendations from ALTO.
o Make (ALTO recommended) data/forwarding plane decisions based on
the cost maps obtained from ALTO.
5.2. Information Flow of Applications, SDN and ALTO
We now give three examples to describe a complete work flow, which
connects all key elements in an SDN.
5.2.1. SDN-aware Applications
o An application's end point sends a request for network resources
to the SDN controller it belongs to (i.e., the SDN controller for
the SDN domain where this application's end point belongs to).
The request should include the destination end point or the set of
destination end points, and a set of requirements on network
resources (e.g., bandwidth)
o The SDN controller obtains an SDN-specific cost map from the ALTO
server (this step may occur independent of remaining steps)
Xie, et al. Expires December 29, 2012 [Page 16]
Internet-Draft ALTO Use Cases For SDN June 2012
o The SDN controller uses the cost map and negotiate one or many
path(s) with other SDN controllers (since the path may span across
multiple SDN domains, thus all SDN controllers of the involved
domains should participate in setting up the paths)
o The SDN controller responds to the requesting application's end
point.
o If the requested path(s) are successfully set up, the
application's end point starts to communicate with the destination
end points.
5.2.2. SDN-unaware Applications
SDN-unaware applications do not directly communicate with SDN
controllers. Instead, they follow special packet formatting rules to
encode the SDN-specific requests, and the SDN capable networking
devices pick up these requests and forward them the SDN controllers.
The remaining work flow is similar to the work flow of SDN-aware
applications, except that SDN controllers do not respond to the
requesting applications. Thus, when the requests cannot be
satisfied, SDN-unaware applications may suffer from packet losses,
due to networking devices process these applications' packets in a
best effort fashion.
5.2.3. Legacy Applications
Legacy applications can be greatly simplified, as it is unnecessary
and is not helpful for them to directly communicate with ALTO servers
any more:
o An end point of a legacy application sends a packet to a known
destination
o A SDN capable networking device picks up the packet; however, if
the path for the two end points has not been set up yet, the SDN
controller will be consulted
o The SDN controller obtains a cost map from the ALTO server (this
step may occur independent of remaining steps).
o The SDN controller negotiate with other SDN controllers to set up
a best-effort path for the requesting end point.
o The forwarding rules for this path are pushed to all networking
devices that are on the chosen path
Xie, et al. Expires December 29, 2012 [Page 17]
Internet-Draft ALTO Use Cases For SDN June 2012
o Communications between the two end points continue; the forwarding
rules may expire if the communication is tore down
In this case, legacy applications are relieved from the complexity of
dealing with the ALTO server using the ALTO protocol. ALTO-related
intelligence, which fundamentally belongs to the network
intelligence, is implemented in the network, rather than partly
outside the network.
5.3. Summary
It is worth noting that this architecture is fundamentally different
from common ALTO use cases such as ALTO in CDN or data center (DC).
The differences lie in that in the latter cases the components in
question (e.g., CDN or DC) are largely consumers of ALTO services,
while in the former case SDN domains are not only making decisions
that may affect ALTO and generating/aggregating information that ALTO
needs, but also the consumers of ALTO services. Furthermore, in the
former case, SDN domains are an integral part of the underlying
network infrastructure where their decisions could be treated as
constraints for ALTO; however, in the latter cases, the components in
question (e.g., CDN or DC) are apparently not necessarily integral
parts of the underlying network and their decisions could be treated
as recommended outcomes suggested by ALTO.
6. Use Case for Upward Flow
The upward flow delivers SDN domains' network information by SDN
controllers to the ALTO server. Each SDN controller is responsible
for collecting, aggregating, and submitting its own domain's network
information to the ALTO server. Due to the possibility of some SDN
domain being dedicated to certain customers, we illustrate the upward
flow in two use cases.
6.1. Unrestrictive Information Exporting
SDN domain controllers have to collect various network information
from the networking devices they manage no matter if ALTO exists or
not. The reason is that an SDN controller may have to make decisions
for allocating resources within its domain, and making such decisions
need various network information. Since such information is readily
collected and available, an SDN controller could submit such
information as is (or after simple processing) to the ALTO
server.Take the available link bandwidth as an example (available
link bandwidth could be used as a measure of "distance"). An SDN
controller could periodically collect the available bandwidth on all
links in its domain and submit it to the ALTO server. However, such
Xie, et al. Expires December 29, 2012 [Page 18]
Internet-Draft ALTO Use Cases For SDN June 2012
information should be annotated with the domain information (e.g.,
domain ID). By submitting such information, later other SDN
controllers may request for this domain's available link bandwidth
information.
6.2. Restrictive Information Exporting
An SDN domain belonging to a carrier may be dedicated to certain
customers of that carrier. In this case, the dedicated users of an
SDN domain manage not only how resources should be allocated but also
what information should be exported.
A carrier may dedicate a set of small data centers (on multiple
sites) to its certain customer. These data centers are put under a
single SDN domain. The customer can manage the dedicated multi-site,
small data centers via the SDN controller. Periodically the SDN
controller collects network information from all data centers.
However, different than the former unrestrictive case, the customer
may have stringent privacy policies and therefore decide to aggregate
the collected information before submitting to the ALTO server.
For instance, the customer may aggregate the information for a data
center network in the same site such that the data center network is
shrunk into a single node; by doing so, the multi-site data center
network is aggregated into a multi-node network topology, each node
in the topology actually corresponds to a small data center in
reality. The aggregated network topology could be annotated with
available link bandwidth information or other information that is
collected and allowed to be exported.
The customer's information aggregation policy defines how the
information should be pre-processed before exporting to the ALTO
server. The main purpose of aggregation is to protect privacy. As a
result of information aggregation, the exported network information
could be a logical topology (annotated with various network
information, e.g., distance or cost) which is totally different from
the physical topology.
6.3. Information Aggregation
Without SDN, ALTO defines cost maps for an aggregated view of the
network topology, where the granularity of aggregation is determined
by the network carrier and could be either coarse-grain or fine-
grain.
However, with the introduction of SDN, such information aggregation
could be greatly simplified and should be policed based on the
Xie, et al. Expires December 29, 2012 [Page 19]
Internet-Draft ALTO Use Cases For SDN June 2012
policies defined for each SDN domain. For instance, ALTO only needs
to collect information from a pre-defined set of SDN domain
controllers, where the controllers determines at what granularity
they would like to aggregate the information and export them. In
addition, such aggregation is governed by the domain-specific
policies, which defines not only the granularity of aggregation but
also to whom such aggregated information may be exposed.
More specifically, an illustrative use case is as follows. SDN
controllers collect fine-grain information and aggregate it
periodically per their policies. ALTO is configured to obtain the
aggregated information from a set of SDN domain controllers and
obtain possibly raw information from networking devices (or the
network operation center). ALTO then constructs a complete view of
the overall network (an aggregated view of the network). SDN
controllers obtain cost maps from ALTO and apply such maps when
making data/forwarding plane decisions.
Another illustrative use case is as follows. SDN controllers may
choose to export fine-grain information to ALTO. After it obtains
the cost maps from ALTO, it could leverage the cost maps with greater
details about their own domains and make informed decisions.
However, SDN controllers should not overload ALTO by exporting too
much fine-grain information.
7. Use Case for Downward Flow
We illustrate the use of downward flow through several use cases as
follows.
Note that when the originating SDN domain's controller make decisions
for choosing path(s) and set up the path(s), each involved SDN domain
controller should map the overall decision to scoped decisions
specifically for their responsible domains.
7.1. SDN-Aware QoS Metrics
We use two use cases to describe SDN-aware QoS. When aggregating QoS
information, SDN controllers or the information aggregation policies
should understand the semantics of each QoS metrics. For instance,
some metrics (e.g., delay) are additive, while some others are
multiplicative (e.g., packet loss rate). The information aggregation
policy should be flexible enough to specify such details.
Xie, et al. Expires December 29, 2012 [Page 20]
Internet-Draft ALTO Use Cases For SDN June 2012
7.1.1. Use Case: On-Demand Bandwidth
An SDN capable application / source end-point may request for a
certain amount of end-to-end bandwidth to a destination end-point on
the fly. The two end points in question should be in the same
administrative domain, but they are not in the same SDN domain. The
path(s) set up for such a request span across multiple SDN domains.
The SDN controller of the source domain (i.e., the SDN domain where
the source end-point is located), referred to as the source SDN
controller, should first obtain the cost maps from the ALTO server.
Such cost maps are SDN-domain-specific, namely, the costs are defined
for pairs of SDN domains, rather than for pairs of end points as in
the legacy ALTO case.
The source SDN domain controller should then determine path(s) for
the two end points based on the cost maps and associated information
obtained from ALTO. More specifically, the controller should:
o Compute a lowest-cost path at the SDN domain level using the
obtained SDN-domain-specific cost map.
o Contact the controllers of those SDN domains on the selected path,
probing for the available bandwidth that could be dedicated to the
requested session.
o Check if all of the selected path have sufficient combined
bandwidth that matches the required bandwidth
o if the combined bandwidth of all selected paths cannot match the
requirement, then go back to step 1 and select another lowest-cost
path (different than the already selected ones)
* if no path can be selected and the combined bandwidth does not
match the requirement, the request cannot be satisfied.
o if the combined bandwidth of all selected paths match the
requirement, then set up all selected paths by signaling all
involved SDN domain controllers. Note that the signaling protocol
and how to set up paths are beyond the scope of this document.
Data backup and migration among data centers, which typically require
bulk data transfers, is an example of on-demand bandwidth use case.
Data centers may be managed by one or multiple SDN domains; thus bulk
data transfer could be thought of as bulk data transfer among
multiple SDN domains.
Xie, et al. Expires December 29, 2012 [Page 21]
Internet-Draft ALTO Use Cases For SDN June 2012
7.1.2. Delay
Similar to the preceding use case, applications may request for paths
satisfying some certain QoS metrics, e.g., VoIP applications may ask
for paths with delay being lower than certain thresholds. This
requires that ALTO cost maps embed such information, and that SDN
controller should export such information to ALTO.
7.2. Content Delivery Networks (CDN)
Content Delivery Network (CDN) has been widely deployed to help
dissemination of content at the Internet scale. Network carries are
also deploying CDNs inside their own networks to improve the user
experiences of their customers. With the introduction of SDN, not
only legacy CDN but also a new SDN-based CDN can be seamlessly
implemented and integrated with the current network infrastructure.
Here is an example of the flow of SDN-enabled CDN. Suppose that
there are a set of CDN servers deployed in a carrier's network and
they are willing to be managed by SDN. An equivalent class for each
of the CDN server is defined by either the CDN carrier or the network
carrier (these two carriers can be the same). An equivalent class is
a set of IP addresses, one for a CDN server, where if one can be used
to fulfill requests for a specific content, then any server in this
class can also be used to serve the same requests. In the extreme
case, there is only one equivalent class for all CDN servers.
Then the pre-defined equivalent classes are pushed to the SDN
controllers, which leverage such information to select CDN servers
and set up paths for any end point to any such servers.
o A network client (e.g., an HTTP-based Web client) obtains the IP
address, referred to as A, of one of the CDN servers in the
carrier's network (e.g., by DNS queries)
o The client sends a first packet destined for A (for HTTP requests,
this packet is a TCP SYN packet)
o An SDN capable networking device picks up the packet
o If there are forwarding rules already set up for the communication
between the requesting client and the destination A, then follow
the rules to forward the request packet
o Otherwise, forward the request packet to the SDN controller of
this domain
Xie, et al. Expires December 29, 2012 [Page 22]
Internet-Draft ALTO Use Cases For SDN June 2012
o Once receiving a forwarded packet from a networking device, the
SDN controller takes the following actions:
* Retrieves the equivalent class for the given destination A
* Obtains a cost map from the ALTO server (this step could take
place asynchronously)
* Ranks all CDN servers in the equivalent class according to the
cost map obtained from the ALTO server
* Selects the best CDN server, referred to as B, based on the
above ranking
* Negotiates and sets up a best-effort path to the selected CDN
server with other controllers
* Sets up forwarding rules for the path, and rewriting rules for
replacing the IP address of A with the IP address of B (so that
the client is actually communicating with B, although it may
think that it is communicating with A; however, which server it
communicates is not important)
o The request packet is forwarded to the chosen CDN server B,
subject to the forwarding rules and rewriting rules
o The client communicates with the CDN server B
o The path and associated forwarding/rewriting rules are expired whe
n the communication is torn down (this step is irrelevant to the
ALTO extension for SDN, therefore, it is out of scope)
However, the above use case has two limitations. First, it violates
the TCP semantics; namely, the client intends to and believes that it
is communicating with server A, but actually it is communicating with
server B. Second, it has to rely on the capability of devices being
able to rewrite forwarding rules (e.g., use one IP address to replace
another one in a packet).
If the above two limitations become concerns, e.g., either TCP
semantics should not be violated or rewriting is not available or
both, the above SDN-enabled CDN use case can be implemented in
similar way, with the help of a redirection server.
Below we describe the steps that are different:
o A redirection server (or server farm), referred to as R, is set up
for redirecting client requests
Xie, et al. Expires December 29, 2012 [Page 23]
Internet-Draft ALTO Use Cases For SDN June 2012
o Each SDN controller sets up path(s) to the given redirection
server R
o Note that the redirection server could be an integral component of
an SDN controller (either collocated or integrated), in which
path(s) are not necessary
o Once receiving a forwarded packet from a networking device, the
SDN controller takes the following actions:
* Retrieves the equivalent class for the given destination A
* Obtains a cost map from the ALTO server (this step could take
place asynchronously)
* Ranks all CDN servers in the equivalent class according to the
cost map obtained from the ALTO server
* Selects the best CDN server, referred to as B, based on the
above ranking
* Sends the information of the chosen CDN server, i.e., its IP
address B, to the redirection server R
* Negotiates and sets up a best-effort path to the redirection
server R (if R is not integrated with the SDN controller)
* Sets up forwarding rules for the path to R
* Negotiates and sets up a best-effort path to the CDN server B
* Sets up forwarding rules for the path to B
o The client communicates with the redirection server R
o R sends an HTTP redirection packet to the client, redirecting
future requests to the CDN server B (which is notified by the SDN
controller)
o The client communicates with the chosen CDN server B (note that
the path to B has been already set up)
7.3. Information-Centric Content Delivery Networks (IC-CDN)
Information-Centric Networking (ICN) is a "host-to-information"
communication model, different from the legacy "host-to-host" model
implemented by the Internet. Content Delivery Network (CDN) is more
of a "host-to-information" model (i.e., CDNs can be treated as a
Xie, et al. Expires December 29, 2012 [Page 24]
Internet-Draft ALTO Use Cases For SDN June 2012
special instance of ICN), but implemented in the "host-to-host"
model, due to the fact that the current semantics provided by the
Internet only support the "host-to-host" model.
With the introduction of SDN, CDNs can be converted into an
information-centric networking implementation:
o A CDN client sends a request for a specific content
o The request packet is formatted per the CDN in SDN specification
(beyond the scope of this draft), containing
* the requested content name in the packet
* destination (a specific anycast IP address) which is reserved
for legacy applications to invoke SDN capabilities
* (optional) QoS requirements (e.g., prefer fast/local servers
vs. slow/remote servers, demands are elastic or not)
o An SDN capable networking device picks up the request packet
o If there are forwarding rules set up for this content request
already, then follow the rules to forward the request packet, and
terminate this.
o Otherwise, forward the request packet to the SDN controller for
this domain.
o The SDN controller communicates with the CDN's name directory to
look up possible CDN servers that can satisfy the request
o The SDN controller obtains a cost map from the ALTO server
o The SDN controller applies the cost map to select the best CDN
server per the QoS requirements if specified in the request
o The SDN controller negotiate the path to the selected CDN server
with other controllers
o The SDN controllers that along the chosen path set up the path,
and push the forwarding rule(s) for this chosen path to all
networking devices that are involved
o The request packet is forwarded to the chosen CDN server
o Data packets flow back to the CDN client
Xie, et al. Expires December 29, 2012 [Page 25]
Internet-Draft ALTO Use Cases For SDN June 2012
In this use case, the CDN clients could be modified to send the
"information-centric" request. However, in a realistic
implementation, neither the CDN clients nor the CDN servers have to
be significantly modified (e.g., CDN redirection could be leveraged
to implement the above work flow).
8. Conclusions
In this draft, we identify the fundamental differences between legacy
ALTO client/server and ALTO client/server with the existence of SDN.
The introduction of SDN fundamentally changes the way that the ALTO
works. We present the Vertical Architecture and the Horizontal
Architecture to allow coherent coexistence of SDN and ALTO. We
believe that the Vertical Architecture allows better division,
management, flexibility, privacy control and long-term evolution of
the network. Therefore we mainly focus on the Vertical Architecture
in this draft. We also define the main interactions and information
flows, and present a set of use cases to illustrate how we extend
ALTO to support SDN, in the Vertical Architecture.
9. Contributors
The authors would like to thank Vijay K. Gurbani for his many
detailed reviews and helpful assistance on this draft.
Vijay K. Gurbani
Bell Laboratories, Alcatel-Lucent
1960 Lucent Lane, Rm. 9C-533
Naperville, IL 60566
USA
Email: vkg AT (acm.org,bell-labs.com)
10. Acknowlegements
The authors would like to thank Aditi Vira for editing the draft.
11. Security Considerations
TBD.
Xie, et al. Expires December 29, 2012 [Page 26]
Internet-Draft ALTO Use Cases For SDN June 2012
12. IANA Considerations
This document makes no specific request of IANA.
Note to RFC Editor: this section may be removed on publication as an
RFC.
13. Informative References
[1] Koponen, T., Casado, M., Gude, N., Stribling, J., Poutievski,
L., Zhu, M., Ramanathan, R., Iwata, Y., Inoue, H., Hama, T., and
S. Shenker, "Onix: A Distributed Control Platform for Large-
scale Production Networks", Proceedings of the 9th USENIX
Symposium on Operating Systems Design and Implementation (OSDI
10), Vancouver, Canada, pp. 351-364", October 2010.
[2] Gurbani, V., Scharf, M., Lakshman, T., and V. Hilt, "Abstracting
network state in Software Defined Networks (SDN) for rendezvous
services, IEEE International Conference on Communications (ICC)
Workshop on Software Defined Networks (SDN)", June 2012.
Authors' Addresses
Haiyong Xie
Huawei & USTC
2330 Central Expy
Santa Clara, CA 95050
USA
Phone:
Email: Haiyong.xie@huawei.com
Tina Tsou
Huawei (USA)
2330 Central Expy
Santa Clara, CA 95050
USA
Phone:
Email: Tina.Tsou.Zouting@huawei.com
Xie, et al. Expires December 29, 2012 [Page 27]
Internet-Draft ALTO Use Cases For SDN June 2012
Diego R. Lopez
Telefonica I+D
Don Ramon de la Cruz, 84
Madrid, 28006
Spain
Phone:
Email: diego@tid.es
Hongtao Yin
Huawei (USA)
2330 Central Expy
Santa Clara, CA 95050
USA
Phone:
Email: Hongtao.yin@huawei.com
Xie, et al. Expires December 29, 2012 [Page 28]