RE: pilc minutes (corrected)
"Farid Khafizov" <faridk@nortelnetworks.com> Thu, 17 January 2002 23:17 UTC
Delivery-Date: Mon Jan 28 09:04:15 2002
Delivery-Date: Thu Jan 17 18:18:47 2002
Delivery-Date: Thu, 17 Jan 2002 18:18:47 -0500
Message-Id: <EF1056F8EB4ED511B8FB0002A56079D40105D1EC@zrc2c014.us.nortel.com>
From: Farid Khafizov <faridk@nortelnetworks.com>
To: "'mallman@grc.nasa.gov'" <mallman@grc.nasa.gov>, 'Phil Karn' <karn@ka9q.net>, 'Aaron Falk' <falk@ISI.EDU>, pilc <pilc@grc.nasa.gov>
Cc: Mehmet Yavuz <myavuz@nortelnetworks.com>, Farid Khafizov <faridk@nortelnetworks.com>
Subject: RE: pilc minutes (corrected)
Date: Thu, 17 Jan 2002 17:17:57 -0600
MIME-Version: 1.0
X-Mailer: Internet Mail Service (5.5.2653.19)
Content-Type: multipart/alternative; boundary="----_=_NextPart_001_01C19FAD.32DBF8C0"
Status: RO
Content-Length: 23061
Lines: 616
Enclosed please find suggested text on Bandwidth Oscillation for LINK-ID and comments on the latest version of the draft. --Farid //------------------------------------------------------------------------- // Suggested addition to LINK-ID (draft-ietf-pilc-link-design-08.txt) // To follow directly "BoD Subnets" section //------------------------------------------------------------------------- Bandwidth Oscillation When shared resources are periodically assigned to users for a limited time, BoD systems may experience Bandwidth Oscillation. One example is a CDMA based wireless data network such as UMTS or CDMA2000. Limited pool of orthogonal codes and RF transmit power necessitates time-sharing of those resources. If a number of users require, for example, large data files transfer at the same time, the system (e.g., the scheduler) may have to repeatedly allocate and de-allocate resources for each user. Such mode of operation causes Bandwidth Oscillation, which can result in spurious timeouts, and, hence, throughput degradation. Although effects of spurious retransmission have been extensively studied in the literature (e.g., [LK00]), it is important to note that in some cases Bandwidth Oscillation can be the single most important factor in throughput reduction. It has been shown that for some configurations of CDMA2000 networks, throughput degradation can be in excess of 50% [n1]. Experiments have also shown that effects of Bandwidth Oscillation can amplify throughput degradation caused by TCP segment loss or reordering. At the time of writing of this document IETF was working on recommendations for TCP options [n2] that could reduce adverse effects of BOS. Consensus of this forum, however, is that a better solution can be found by properly designing sub-networks. For a fixed TCP configuration, throughput reduction depends on the high bandwidth (HB), the duration of HB (B), the low bandwidth (LB) and the duration of LB (D). By choosing appropriate combination of <HB,B,LB,D> throughput degradation can be reduced or even avoided. However, in cases when multiple users share same resources, B and D are related and reducing D becomes impossible without reduction of B. In addition, attempts to reduce B and/or D may increase signaling overhead, and, therefore, limit designers' flexibility in choosing the optimal configuration. [n1] F.Khafizov, M.Yavuz, "Running TCP over IS-2000", to appear in Proc. of IEEE ICC 2002 [n2] H. Inamura, G. Montenegro, M. Hata, M. Hara, J. James, W. Gilliam, A. Hameed, R. Ludwig, R. Garces, P. Ford, F. Wills, "TCP over 2.5G and 3G Wireless Networks", work in progress as internet-draft, draft-ietf-pilc-2.5g3g-05, November 2001 //------------------------------------------------------------------------- // Comments on existing text available at http://people.qualcomm.com/karn/pilc.txt //------------------------------------------------------------------------- Section: Choosing the MTU in Slow Networks > "However, the link resources used to > send the aborted packet are lost, and overall throughput will > decrease." It is probably me not being a native speaker, but I felt the above sentence wasn't very clear: I had trouble with present and future tense verbs in one sentence. Perhaps it should be written as "However, when the link resources used to send the aborted packet are lost, the overall throughput decreases." or "However, in this case, the link resources used to send the aborted packet are lost, and overall throughput decreases." //------------------------------------------------------------------------- Section: How TCP Works > "Since the most common cause of packet loss is congestion" Although this is true considering overall Internet traffic, in (at least some) wireless networks this assumption is not true. Experiments suggest that packet loss due to high BER is comparable or higher than that due to congestion. Perhaps that fact should be mentioned with the reference to rfc3155. //------------------------------------------------------------------------- Section: How TCP Works > "Recent proposals have been made to lower the > dupack threshold to 2." Would be nice to have a reference to "proposals" //------------------------------------------------------------------------- Section: TCP Performance Characteristics -> Analysis of Link-Layer Effects on TCP Performance > "If this link layer were used in the Internet, on a path that > otherwise had a round trip of of 80 msec, " "of of" //------------------------------------------------------------------------- Section: Bandwidth Asymmetries > "acknowledgments is too large, the slow return of of the ACKs directly" "of of" //------------------------------------------------------------------------- Section: Buffering, flow & congestion control > "Available Bit rate (ABR" should have "Rate" //------------------------------------------------------------------------- Section: Compression > "We make a stronger recommendation that subnetworks operating at low > speed or with small MTUs compress IP and transport-level headers" should be "We make a stronger recommendation that subnetworks operating at low BER conditions and low speed or with small MTUs compress IP and transport-level headers " It is mentioned later in the text that in the presence of errors header compression is not a good idea. Here one might add that in such circumstances there is no benefit of Fast Retransmit/Fast Recovery mechanism. //------------------------------------------------------------------------- Section: Mobility > "the most appropriate and efficient. Mobility is already an feature of" article change "an" -> "a" //------------------------------------------------------------------------- Section: Security Considerations > "2. Use of 'security by obscurity' [Schneier4] [Crypto9912]. One" and > "3. Inclusion of trapdoors [Schneier4] [Crypto9912]. Trapdoors are" Reference [Schneier4] is not defined //------------------------------------------------------------------------- > -----Original Message----- > From: Aaron Falk [SMTP:falk@ISI.EDU] > Sent: Monday, January 14, 2002 1:04 PM > To: pilc; minutes@ietf.org > Subject: pilc minutes (corrected) > ========================= > > CDMA2000 > > F. Khafizov presented slides on TCP performance over CDMA 2000 > networks. Issues to be addressed include bandwidth oscillation, high > BER, and bandwidth asymmetry. Bandwidth oscillation degrades > throughput through spurious timeouts. True of any system with widely > varying bandwidth. > > * Bandwidth Oscillations (BO) > - Significant throughput degradation > - Not cdma2000 specific > - Related to resource contention > - Fixes include 1) fix TCP, 2) Fix network or 3) leave as is. > - Possible fixes > o Adjust RTO value > o enable Time Stamp option > o increase window size > o minimize ACK delay > - Other recommendations > o Aggressive link level error recovery > o disable VJ header compression > o use SACK > - Options > o Incorporate BO in 2.5g/3g and/or LINK ID > o make a separate ID for BO > o have a section in 2.5G/#g ID addressing specifics of CMDA2000 > > Suggestions from chair and floor moved toward including in existing > documents rather than stand-alone document. > > F. Khafizov volunteer to generate text for the documents. > > >
- pilc minutes (better late than never....) Aaron Falk
- pilc minutes (corrected) Aaron Falk
- RE: pilc minutes (corrected) Farid Khafizov