Internet DRAFT - draft-mcdysan-iccrg-usecases
draft-mcdysan-iccrg-usecases
Conex Group D. McDysan
Internet Draft Verizon
Intended Status: Informational
Expires: September 5, 2012
March 5, 2012
Potential Use Cases of Internet Congestion Control Research
draft-mcdysan-iccrg-usecases-00.txt
Status of this Memo
This Internet-Draft is submitted to IETF in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
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."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html
This Internet-Draft will expire on April 17, 2011.
Copyright Notice
Copyright (c) 2012 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. 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.
McDysan Expires September 5, 2012 [Page 1]
Abstract
This draft proposes some use cases for consideration by the iccrg.
These use cases are in addition to and/or complement those described
by the conex working group[UseCases]. The conex working group had
determined that these use cases were out of scope of the current
charter, and/or could potentially be built out of the conex abstract
mechanism. The focus of the use cases in this draft is on forms of
congestion exposure that involve resources other than queues and
timeframes other than real-time. Background and motivation for these
use cases is first discussed. Expanded material on the usage
tier/volume use case is also provided.
Table of Contents
1. Introduction...................................................2
2. Conventions used in this document..............................3
2.1. Acronyms..................................................3
2.2. Terminology...............................................3
3. Motivation and Background......................................4
4. Potential Use Cases for Experimental Research..................5
4.1. Inequity of Heavy versus Light Users......................5
4.2. Feedback on Time of Day, Day of Week Charging.............6
4.3. Recharging for Implementing Congestion Pricing............7
4.4. Usage Tier/ Volume Feedback...............................7
4.4.1. Objectives for Addressing this Issue.................7
4.4.2. Potential Support Using Abstract Mechanism...........8
4.4.3. Additional Support Using other Measures and Mechanisms9
5. Security Considerations.......................................11
6. IANA Considerations...........................................11
7. Acknowledgements..............................................11
8. References....................................................11
8.1. Normative References.....................................11
8.2. Informative References...................................11
9. Acknowledgments...............................................12
1. Introduction
This draft proposes some use cases for inclusion in the conex Working
group charter's deliverable for an informational RFC covering use
case description. These use cases are in addition to and/or
complement those described in [UseCases], and focus on forms of
congestion exposure that involve resources other than queues and
timeframes other than real-time.
Section 3 provides some motivational background and a statement of
problems involved with congestion pricing, with references to the
presentations by experts in this area at the IETF 78 Technical
Plenary in Maastricht.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 2]
Section 4 provides text for each of the above use cases in a
mechanism independent manner.
As requested in the Beijing meeting, this is an individual draft that
expands on the usage tier/volume use case from [McDysan]. The
feedback recorded in the Beijing conex meeting minutes was that a
number of people were interested in this idea, but that it was out of
scope of the current charter, and/or could potentially be built out
of the conex abstract mechanism.
Section 3 provides some motivational background and a statement of
relevant problems involved with congestion pricing.
Section 4 provides text for the usage/volume tier feedback use case.
It contains a section that covers a problem statement, objectives for
resolving this issue, potential approaches for implementing this use
case employing currently defined conex mechanisms, and a description
of additional measures and mechanisms that could solve the stated
problem and issues.
2. Conventions used in this document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC-2119 [RFC2119].
2.1. Acronyms
conex = congestion exposure
2.2. Terminology
The following is a quote from the CONEX working group charter:
" ... develop a mechanism by which senders inform the network about
the congestion encountered by previous packets on the same flow ...
at the IP layer, such that the total level of congestion is visible
to all IP devices along the path"
Despite its central role in network control and management,
congestion is a remarkably hard concept to define. The discussions
in [Bauer09] provide a good academic background. [RFC6077] defines
it as "as a state or condition that occurs when network resources are
overloaded, resulting in impairments for network users as objectively
measured by the probability of loss and/or delay." An economist
might define it as the condition where the utility of a given user
decreases due to an increase in network load. The economic
implications are obviously different for an end user versus a service
provider.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 3]
3. Motivation and Background
This section provides references to relevant presentations given by
experts on congestion pricing from the IETF 78 Technical Plenary in
Maastricht.
Successful adoption of experimental congestion
pricing/exposure/information sharing protocol(s) must address use
cases that provide significant value to users, content providers and
service providers. Central themes to this value proposition are
incentives (i.e., congestion pricing) and the cost of providing
marginal capacity [Varian, Johari].
There are three time scales over which congestion pricing can operate
[Johari]: short (milliseconds to seconds), medium (minutes to hours
to days) and long (months to years). Currently, the short term
congestion signal is lost packets or a specific indication of
congestion of a particular resource (e.g., a ECN indication for queue
congestion), as stated in the conex charter. Setting congestion price
as the marginal value of capacity is useful for medium timescales via
traffic engineering and longer time scales via provisioning [Johari].
Some congestion exposure problems are challenging to address and are
not completely addressed in conex [UseCases]:
o 20% of the users generate 80% of the traffic and create
unfairness with certain resource sharing [Varian]
o Volume-based pricing makes it difficult for users to manage costs
incurred, [Varian]. Note that a significant number of ISPs
implement some form of usage/volume caps in at least some parts of
the world [NewAmerica].
o Customers will pay a premium for unmetered use [Varian]
o A form of congestion pricing is "recharging" (e.g., "free
shipping") [Varian], where someone other than the end user pays
for incurred congestion.
o In general a content or service provider has hard capacity
constraints at certain bottlenecks in their infrastructure (e.g.,
server capacity, router interface/queue service rate). Some form
of adaption, such as time-shifting, route-shifting, or moderating
the demand is required to adapt to these constraints [Kelly].
If conex exposes congestion without damage (e.g., loss) then many
forms of adaption are feasible, as long as incentives are aligned
with the signaled congestion [Kelly]. The use cases in the next
section focus on forms of adaption that enable certain sets of
incentives that are not completely covered in [UseCases].
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 4]
4. Potential Use Cases for Experimental Research
The following use cases have been deemed to be out of scope for the
current conex wg charter. They may be candidates for specific areas
of research
4.1. Inequity of Heavy versus Light Users
In many networks, 20% are heavy users generating 80% of the traffic
[Varian]. This means that a heavy user generates 16 times the traffic
as a light user as a medium term (e.g, monthly) average. But, in a
bandwidth-tiered flat priced network, heavy and light users often pay
nearly the same price since pricing is based upon the short-term
(milliseconds) bandwidth measure of a shaper and/or a policer.
During non-peak periods, the resources of a service or content
provider are underutilized and the marginal cost of capacity is
small.
The access network is provisioned and traffic engineered for peak
capacity of all users, and when congested, heavy users create 16
times the congestion of small users.
However, during peak use periods, a heavy user may send at near the
bandwidth tier while light users may send intermittently. There is a
need for a means for service providers to equitably assign costs to
heavy versus light users. For example, the light users may pay less
if they were charged by volume, as described in another use case.
A congestion measure of burstiness (e.g., ratio of peak rate to
average rate over a longer interval than the tiered bandwidth shaper
or policer) could be helpful in this use case. In general, a bursty
packet flow is light (e.g., web surfing) and a non-bursty packet flow
is heavy (e.g., viewing a lengthy HD video). The destination could
perform this measure and feed it back to the sender. It may only be
necessary to feedback this measure for heavy users, since the absence
of such feedback could be inferred as an indication of a light user.
The sender could insert some processed version of this feedback
measure this at the IP layer so that all IP devices could be aware of
whether this is a heavy or light user.
The following use case is an example of how the conex indication of
congestion may be useful to handle the heavy versus light user case.
The following was based upon discussions with Toby Moncaster.
Over a time period related to the statistical multiplexing or
economic congestion interval (e.g., many seconds to minutes to hours)
total up the number of bytes that have been congestion marked and the
total number of bytes sent per end-user. Compute the ratio of
congested bytes to total bytes. This measures the average rate per
user.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 5]
Quantizing users into classes using one threshold on total and
another threshold on ratio results in a grid that identifies four
classes of user:
+------------+-------------+-------------+
| | Volume |
| Ratio | Large | Small |
+------------+-------------+-------------+
| High | Heavy User | Bursty User |
+------------+-------------+-------------+
| Low | LEDBAT User | Light User |
+------------+-------------+-------------+
(Where "LEDBAT User" includes other Less-than-Best-Effort
algorithms.)
Figure 1: Four Classes of User
Note that Bursty and Heavy Users contribute more to congestion
marking, but a Bursty user contributes less overall congestion
marking and may be creating shorter periods of queue filling as
compared with heavy users. LEDBAT and light users create less to
congestion marking, with LEDBAT users able to transfer more volume as
compared with light users since LEDBAT users back off before
congestion marking occurs. An operator might reasonably take this
into account in their shaping algorithms.
4.2. Feedback on Time of Day, Day of Week Charging
Congestion occurs when the offered load approaches that of the
provisioned capacity, which often does not occur until shortly before
there would be a need to provision additional capacity. Depending
upon how restoration capacity is allocated by a service/content
provider, congestion may only occur during peak periods when a
failure is present.
Without Conex, utilization averaged over several minutes can be as
high as 70 to 80% in typical network bottlenecks without loss that
would reduce TCP effective throughput (i.e., goodput). A short term
congestion control (sub-second to seconds) method that could increase
utilization to above 90% and still avoid loss would only increase
effective capacity by 10-20%.
If traffic increases at 50-75% per year, then a 10-20% increase in
effective capacity means that the provisioning interval is only
shifted by a few months. This handling of short term congestion use
case alone may not be sufficient motivation for a service provider to
deploy congestion handling measures.
However, in many points of a network, the majority of usage occurs
during peak periods (e.g, a few busy hours) while much spare capacity
exists off peak. The product of the spare capacity (bits/second) and
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 6]
the non-peak interval (seconds) that could carry traffic ranges from
2 to 10 times of the traffic carried during the peak period.
Congestion exposure and congestion pricing that enables users and
content providers to time shift traffic to off-peak periods that
would have otherwise been sent during peak periods can reduce
provisioned capacity cost by as much as several hundred percent.
Historical time of day usage patterns could be employed to time shift
traffic, but often maintenance actions are performed during the off
peak periods, making the prediction of congestion using these methods
less reliable. An automatic method for detection of congestion during
off-peak periods is highly desirable.
4.3. Recharging for Implementing Congestion Pricing
There should be a means to recharge (i.e., someone other than the
receiving user pays) for usage that causes congestion during peak
demand period versus that which does not [Varian].
If TCP were augmented with information related to the form of
congestion, including not only short term as covered in [UseCases],
but also including usage tier, Time of Day, or burstiness then
sufficient information to implement sender pays (e.g., Content
provider)versus receiver pays (e.g., end user) could be implemented.
When the sender includes this information in the IP layer, then usage
tier counting and TOD counting could be accounted for differently in
IP devices in the path between sender and destination. Such an
indication of recharging would need be to be authenticated in some
way.
4.4. Usage Tier/ Volume Feedback
Usage/volume caps may be arranged into multiple tiers with different
pricing based upon monthly volume. This results in some problems as
described in the next section. Next, high-level objectives to address
these issues are then proposed. Then potential ways that already
defined means [Mechanisms] may be employed are described. Finally, to
address some of these issues, other measures and mechanisms that
could possibly better meet the objectives are described.
4.4.1. Objectives for Addressing this Issue
Provide a way to inform a receiver of the usage/volume incurred to a
moment in time. Ideally, this would also include the usage/volume
time period (e.g., a month). This should be accomplished without
having the user logging in to a portal to retrieve usage information,
for example, as described in [Feamster11], but should be provided as
an interface usable (potentially as an API) to any authorized
application.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 7]
Provide a way to inform a receiver of a trend that if usage continues
at the same rate then a specific usage/volume tier will be crossed.
Indicate to a receiver whether usage/volume counting is occurring in
a different way when congestion measure of a particular form (e.g.,
loss, ECN marking) is occurring.
Standardize a way to mark packets in a way (e.g., [Lower Effort]) in
conjunction with some form of conex signaling that indicates
usage/volume counting will not occur (or are counted separately)
under the condition that these packets do not create congestion. A
means to ensure that these marked packets do not create congestion
and do not impact best (and better) effort marked packets is also
required.
Enable a means for recharging to occur, where usage/volume counting
does not occur for the receiving user since some other party has
agreed to incur the cost of usage/volume for that flow.
4.4.2. Potential Support Using Abstract Mechanism
The conex abstract mechanism [Mechanism] defines implicit signaling
of loss and explicit signaling of ECN marking. It also defines re-
echoed signal for loss and ECN marking based upon feedback carried by
TCP from a receiver back to the sender (in an RFC to be developed by
the conex wg as defined in the charter). If this feedback mechanism
is designed to be extensible, then a variety of forms of feedback
could be developed for use in experiments.
Counting usage/volume differently for congested packets (or congested
intervals) based upon [Mechanism] re-echoed congestion experienced
signals seems straightforward. This could be a local matter for the
IP node which implements usage/volume tier counting.
Also, counting packets marked as Lower Effort differently is a local
matter. How to ensure that these packets do not interfere with best
effort could be implemented by Diffserv methods locally and at other
potential bottlenecks.
Since packets subject to counting in a usage/volume cap may not occur
during congestion intervals, reinsertion of such counting information
using the re-echoed signals that indicate congestion does not seem
possible since the same bits cannot represent usage counting and
congestion experienced.
What is missing from the current conex mechanism is a feed forward
path operating over a longer timescale that contains sufficient
information which can be provided to an individual application to
meet the stated objectives.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 8]
4.4.3. Additional Support Using other Measures and Mechanisms
Usage/volume counting has some aspects similar to that of a
congestible queue, but on a much longer timescale, as follows:
o Instead of a queue which is typically sized for O(10 ms) at the
sending rate, usage/volume counting occurs on a timescale O(month.
o A usage/volume tier is a threshold on a long term usage counter,
similar to the way ECN marking can be a threshold on in a queue.
o Queue loss is similar to a usage/volume counter crossing from one
tier into the next.
o A usage/volume tier trend warning is similar to a rate estimate
for ECN marking based upon queue fill rate, as is described in
PCN.
Therefore, in an abstract way a usage/volume counter can be viewed as
a congestible resource, but in some ways not the same as a
congestible queue. If this information is to be fed forward in a way
observable at the IP layer and fed back at the transport layer (e.g.,
TCP), then additional packet and transport fields and/or mechanisms
may be better suited to this purpose.
Furthermore, instead of feeding forward information in each IPv6
packet as in [Mechanism], usage/volume congestible resource
information can occur much less frequently (e.g., many minutes to
hours). The following is an outline of such measures and mechanisms.
The basic idea is based on the fact that the sender and receiver need
to be cooperating using the same experimental extensions to TCP, and
that if TCP can carry some of the additional information, then the
scarcity of IPv6 header bits is avoided. Furthermore, as described
previously, "fast path" processing is not required for this use case
which has resulted in conex specifying the destination options field
[conexdestopt]. Instead, the hop-by-options field of the IPv6 header
could be used [RFC2460], [IPv6Format] which would only require "slow
path" processing. A mechanism to allow the experimental sender to
send a "probe" in the IPv6 packet (e.g., using an experimental IPv6
protocol type) could be used by a intermediate IP node(s) to forward
the "probe" packet to a special processor (which may be separate from
the routers' processor). This special processor could use a polled
version of usage/volume count information per user and could also be
configured with subscription information (e.g., usage/volume cap
tier, cap duration), and threshold settings.
The handling of this "probe" IPv6 packet and associated TCP segment
needs to be done within the TCP flow. It could use an Out Of Band
mechanism similar to the urgent data capability in TCP. (For example,
an experimental usage of the Urgent bit could possibly be employed.)
The special processor could insert additional measures and implement
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 9]
some of the proposed mechanisms and then modify/augment the "probe"
TCP (urgent-like) segment with the requested information and forwards
this modified "probe" packet toward the receiver via the intermediate
IP node. Packets from the receiver back to the sender could be sent
directly, or could be directed through the special processor at
intermediate node(s), depending upon the specifics of the use case
involved.
A consequence of the above extension of measures and mechanisms is
that the sender and receiver now have much more information which
could be used to solve the stated problems and meet the objectives.
The information carried in a "probe" TCP segment could include:
o The service being requested, for example:
o Request information on the users' usage/volume tier
o Request statistics on usage
o Request threshold trend report
o Request not counting this flow since it is lower effort
o Request recharging
o Information that could be provided by the "special processor"
includes:
o Duration and cap for the usage volume measurement tier (e.g., a
month)
o The absolute count of packets and octets received/sent, and/or
fraction of the usage tier already used
o Count of packets and octets received/sent which experienced
congestion
o Count of packet and octets received/sent that were marked as
Lower Effort
o Estimate of whether the user will exceed the usage tier if the
historical usage rate to the reporting instant continues
o A pointer (e.g., URL) and identification of the authentication
method that would enable other queries, and/or implement
alternative charging methods (e.g., recharging)
o Other measures related to the "congestion" of a usage/volume
tier use case (or possibly other use cases as well).
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 10]
One example of a different type of measure is described in
[Stanojevic]. In this paper, the Shapley value [Shapley] is used
instead of a 95-th percentile measure of hourly usage measurements
across a month. The Shapley value has the following desirable
intuitive properties [Shapley]: individual fairness, efficiency,
symmetry and additivity. Although the 95-th percentile measure is not
directly related to the usage/volume tier use case, the authors state
that this is a case they plan to address in future research. A
mapping, such as an approximation to the Shapley value described in
this paper, could be a way to compress the usage/volume tier feed
forward/ feedback information into a smaller number of bits that
represents the incentives described in the objectives section.
5. Security Considerations
In the proposed mechanisms there are indications that could be
spoofed and/or used to game counting and congestion feedback
mechanisms, and therefore an authentication mechanism may be needed
when this information is handled at the TCP/IPv6 layer in the sender
to destination direction or at the TCP layer in the destination to
sender direction.
6. IANA Considerations
None
7. Acknowledgements
The idea of how to use the experimental mechanism being developed by
the conex working group to address heavy versus light users was
suggested by Toby Moncaster. The idea of not counting lower effort
traffic against a usage/volume cap was suggested my Mikael
Abrahamsson on the conex mailing list. Lars Eggert provided the
reference to [Stanojevic] on the conex mailing list as an example of
a different form of congestion measure.
8. References
8.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
8.2. Informative References
[UseCases] B. Briscoe, R. Woundy, A. Cooper, "ConEx Concepts and Use
Cases," draft-conex-concepts-uses-04, Work in Progress
[Bauer09] Bauer, S., Clark, D., and W. Lehr, "The Evolution of
Internet Congestion", 2009.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 11]
[Mechanism] M. Mathis, B. Briscoe, "Congestion Exposure (ConEx)
Concepts and Abstract Mechanism," draft-mathis-conex-abstract-mech-
00, Work in Progress
[Varian] Hal Varian, Google, "Congestion pricing principles," IETF 78
Technical Plenary, 29 July 2010
[Johari] Ramesh Johari, Stanford University, "The information in
congestion prices: milliseconds to years," IETF 78 Technical Plenary,
29 July 2010
[NewAmerica] Li, Losey, "Bandwidth Caps for Residential High- Speed
Internet in the U.S. and Japan," August 2009,
http://www.newamerica.net/files/Bandwidth%20Caps%20for%20High-
Speed%20Internet%20in%20the%20U.S.%20and%20Japan.pdf
[LowerEffort] R. Bless, K. Nichols, K. Wehrle, "A Lower Effort Per-
Domain Behavior (PDB) for Differentiated Services," RFC3662, December
2003
[RFC2460] S. Deering, R. Hinden, "Internet Protocol, Version 6
(IPv6) Specification," RFC 2460 December 1998
[IPv6Format] S. Krishnan, M. Kuehlewind, "Conex IPv6 Format," draft-
krishnan-conex-ipv6
[conexdestopt] S. Krishnan, M. Kuehlewind, C. Ucendo, "IPv6
Destination Option for Conex," draft-ietf-conex-destopt, work in
progress.
[Feamster11] N. Feamster, "Software Defined Network Management," Open
Networking Summit, Oct 2011, http://opennetsummit.org/talks/feamster-
pdf.pdf
[Stanojevic] Stanojevic, Laoutaris, Rodriguez, Telefonica Research,
"On Economic Heavy Hitters: Shapley value analysis of 95th-percentile
pricing," IMC'10, November 1-3, 2010, Melbourne, Australia,
http://conferences.sigcomm.org/imc/2010/papers/p75.pdf
[Shapley] Wikipedia, "Shapley value,"
http://en.wikipedia.org/wiki/Shapley_value
9. Acknowledgments
This document was prepared using 2-Word-v2.0.template.dot.
Copyright (c) 2012 IETF Trust and the persons identified as authors
of the code. All rights reserved.
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 12]
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
This code was derived from IETF RFC [insert RFC number]. Please
reproduce this note if possible.
Authors' Addresses
Dave McDysan
Verizon
Email: dave.mcdysan@verizon.com
mcdysan-iccrg-usecases-00 Expires September 5, 2012 [Page 13]
