[e2e] a question about the deployment of SACK TCP

Kostas Pentikousis kostas at cs.sunysb.edu
Fri Jul 9 14:22:20 PDT 2004


On Fri, 9 Jul 2004, Mark Claypool wrote:

|Ron Lee writes:
| > Only 5% of the TCP connections actually use SACK, (From Anja
| > Feldmann, trace from 12/99) according to

Off-the-box Windows 2000/XP/2003, Solaris 9, and Linux 2.4+
advertize SACK capability. No patches needed.

| > there an up-to-date answer to this? What fraction of the
| > bytes/packets/TCP-flows in the Internet are SACK-capable?

"in the Internet" is quite difficult to say.  Padhye & Floyd used
TBIT in "Identifying the TCP Behavior of Web Servers" to determine
what TCP flavors are deployed "out there". (Ignore for a moment
the fact the study is limited to web servers.) They found that a
large percentage of hosts (more than half) advertize SACK, but
only 6% of hosts use SACK correctly (see also their NANOG
presentation: http://www.nanog.org/mtg-0010/tcp.html)

|Our analysis of some data gathered from a commercial broadband
|provider in Fall 2003 shows about 90% of the flows are SACK-enabled.

We've been studying traces from the NLANR/PMA repository for some
time now.  The first batch of results from 12 different sites (see
"Quantifying the deployment of TCP options - A comparative study",
http://www.cs.stonybrook.edu/~kostas/art/) shows that SACK is
likely to be advertized in the vast majority of TCP connections
(~80%).  Nevertheless, the percentage of "pure ACKs" in the traces
is relatively high, which begs the question: how many SACK blocks
are actually transmitted (let alone taken advantage of)?

|Further analysis suggests that SACK significantly reduces the
|number of retransmissions compared to the non-SACK enabled flows.

This is quite interesting (esp. cf. Balakrishnan et al., "TCP
Behavior of a Busy Internet Server: Analysis and Improvements").

In any case, enabling SACK and getting a benefit from it are two
different things.  Although it is quite possible that nowadays
more than 6% of the hosts use SACK correctly, our results indicate
that SACK is not accompanied by large windows advertisements,
which could significantly reduce the benefits from SACK deployment
"out there".

Best regards,

Kostas
__________________________________________________________________
Kostas Pentikousis                   www.cs.stonybrook.edu/~kostas


More information about the end2end-interest mailing list