2.8.11 Performance Implications of Link Characteristics (pilc)

NOTE: This charter is a snapshot of the 50th IETF Meeting in Minneapolis, Minnesota. It may now be out-of-date. Last Modified: 14-Mar-01


Spencer Dawkins <spencer.dawkins@fnc.fujitsu.com>
Aaron Falk <afalk@panamsat.com>

Transport Area Director(s):

Scott Bradner <sob@harvard.edu>
Allison Mankin <mankin@east.isi.edu>

Transport Area Advisor:

Allison Mankin <mankin@east.isi.edu>

Mailing Lists:

General Discussion:pilc@grc.nasa.gov
To Subscribe: majordomo@grc.nasa.gov
In Body: subscribe pilc
Archive: http://pilc.grc.nasa.gov/pilc/list/archive/

Description of Working Group:

Erik Nordmark (nordmark@eng.sun.com) is the Technical Advisor.

The Internet network-layer and transport-layer protocols are designed to accommodate a very wide range of networking technologies and characteristics. Nevertheless, experience has shown that the particular properties of different network links can have a significant impact on the performance of Internet protocols operating over those links, and on the performance of connections along paths that include such links. This is especially of concern to the wireless networking community.
The PILC working group will produce several BCP/Informational documents. The first document will discuss considerations for link-layer designers from the perspective of best supporting existing IETF protocols will be produced. The next document will discuss the capabilities, limitations and pitfalls of 'performance enhancing proxies' (PEPs), that is, active network elements that modify or splice end-to-end flows in an attempt to enhance the performance they attain in the face of particular link characteristics. The remaining documents will either discuss the impact and mitigations for a problematic link-layer characteristic (or group of closely related characteristics), or provide overviews of which other PILC documents apply to particular problem domains.
As one of its first work items, the WG will review an existing I-D on considerations for "long, thin" networks (one of the salient characteristics of terrestrial wireless links). This will be published as a preliminary assessment of the problem domain, to be refined by later PILC documents.
All documents will identify which of their considerations remain research topics versus which are established as advanced development. Research topics will be explicitly flagged as not part of any recommendations. All documents will also identify any security implications associated with their considerations.
The working group will also serve as a forum for discussing possible modifications to IETF protocols to improve performance in environments with problematic link characteristics - however, not to the detriment of performance and stability in the general Internet, nor to undermine existing security models.
It is incumbent upon the chairs to ensure that the WG maintains good communications with other groups interested in related technology issues, such as wireless forums.

Goals and Milestones:



Submit Internet-Draft on significantly low bandwidth links.



Submit Internet-Draft on significantly lossy links.



Submit Internet-Draft on long-thin networks (based on draft-montenegro-pilc-ltn-01.txt) submitted to the IESG for publication.



Draft of link-layer design considerations document.



Draft of PEP capabilities and limitations document.



Draft on asymmetric network paths.

Oct 99


Draft of TCP Over Wireless document to the IESG as BCP

Nov 99


Document on low bandwidth links to IESG for publication as BCP.

Nov 99


Document on link-layer design considerations submitted for publication as BCP.

Nov 99


Document on lossy links to IESG for publication as BCP.

Nov 99


Document on PEP capabilities and limitations submitted for publication as Informational.

Nov 99


Document on asymmetric network paths submitted to the IESG for publication as BCP.

Nov 99


Possible rechartering of WG to address modifications to IETF protocols.

No Request For Comments

Current Meeting Report

PILC WG Meeting
Thu Mar 22, 2001
Reported by: Spencer Dawkins/Aaron Falk/Mark Allman
Slides available at http://pilc.grc.nasa.gov.

Working Group Status:
ERROR and PEP through WG last call and have been submitted to the AD. The IDs will be put on the IESG plate soon.

SLOW has passed WG last call and will be forwarded to the IESG after this meeting.

LINK is on track for last call (in next few weeks).

Documents still pending: 2.5g3g, ASYM and ARQ. The goal is to conduct WG last calls for all these documents before the next IETF meeting.

The PILC charter calls for possible re-chartering to address some of the issues raised by the group. However, to dat no issues have been raised and therefore PILC will likely go dormant after the current set of documents are completed.

SLOW Status - Markku Kojo
WG last call complete.

A few minor language tweaks are being added:
* MTU size selection text clarified.
* Slow start text clarified.

Document now explicitly recommends Limited Transmit.

LINK Status -- Phil Karn
Lots of typos and nits fixed.

QoS architecture description added (written by Dan Grossman).

ARQ section added, all known issues resolved.

Phil recommends going ahead with WG last call after this meeting.

Issue: should subnetwork encryption be added? Yes, if free, but not a substitute for higher level encryption to protect user data end-to-end.

Issue: some questions raised on list regarding peformance formulae, need experts to resolve issue, editor not the right person.

Reiner Ludwig: on performance formula - holds for connections with high degree of multiplexing, not as applicable for links with few connections, like those outlined in SLOW.

Dan Grossman: on security - we should recommend that link designers bring security of wireless links up to meet expectations users have of point-to-point links.

Markku Kojo: on flow seperation language - concerned about implicitly recommends that it's the LINK layer which is doing the flow seperation, should be IP layer. Link designers may implement in hardware preventing addition of e.g. new DS codepoint.

Reiner Ludwig: doc should only point out that this can be helpful.

Gorry Fairhurst: some text on this in ARQ draft.

Spencer: on formulas - please review forumlas and continue to identify boundry conditions.

Sally: state that forumulas are approximate rules of thumb, not explicit, there are many assumptions on TCP type, traffic characteristics.

Reiner Ludwig: they are just not valid for low mulitplexing connections, need to add language stating thus.

ARQ Status -- Gorry Fairhurst
This draft was originally slated to be rolled into LINK. However the current feeling is that content is rich enough to warrent a seperate document.

See slides for overview of the presentation.

Several outstanding issues:

* sharing & low persistance
* bumps & bursts
* classification of application needs
* multiple links along a path, and duration of high persisitance
* shared channels -- cost of retransmission on a shared channel

Comment from Andre Wutoff (sp?) GPRS-2.5G: have two link layers both capable of retransmission, one of IP packets and one for fragments. Would like to see discussion of performance tradeoffs for persistancy at the two different layers.

Dan: don't think perfect persistancy is real.

Gorry: perfect refers to retransmission up to N2 times a timeout. High is much higher than RTT. Low is on the order of an RTT.

Bill Simpson: use "very high" or "ultra high" rather than "perfect".

Dan: discuss retransmitting fragments vs. whole packets.

ASYM Status -- Gorry Fairhurst

The significant changes from the last rev of ASYM are:

* more examples and recommendations have been added
* a bit about the relationship of this document to PEP was added
* the various techniques have been better organized

The authors are soliciting additional examples of asymmetric networks. In addition, the authors wish any sort of input the comunity has to offer (taxonomy right? recommendations OK? etc.)

Dan Grossman suggested that DSL is a major example.

ludwig: what are the "problem ratios"? misunderstanding about the definition of k in the id, i think

2.5G3G Status -- Hiroshi Inamura
Background info presented and can be found in the I-D.


Gorry Fairhurst: MTU recommendation - does MTU realy have more loss?

Hiroshi: MTU doesn't affect loss rate.

Gorry Fairhurst: name server reference in ID as part of Ricochet - does not refer to DNS and may confuse readers.

???: is this profile just for next generation WAP or general wireless? How would PMTU work in general case?

Hiroshi: by using example of 2.5G/3G you know the MTU.

Gorry Fairhurst: timestamps -- why recommend timestamps? Is this for PAWS? Is there a really high variation of RTT? Do we need timestamps?

Hiroshi: recommendation came from need for window scaling.

Gorry Fairhurst: only need window scaling (and, hence PAWS) for high bandwidth links. Not needed for this application. Should verify need.

Mark Allman: timestamps should not be recommended. Need to see justification if you want to recommend it. PAWS doesn't apply here. As for RTTM, conservative RTT estimate given in RFC 2988 make it unnecesssary.

Carston Borman: ROHC wants to enable timestamps with header compression which should help if you need to use it.


TCP Performance Implications of Network Asymmetry
Link ARQ issues for IP traffic
PILC BCP RFC: TCP Over 2.5G/3G Wireless"