Bug 126678 - poor net performance to certain sites
Summary: poor net performance to certain sites
Keywords:
Status: CLOSED NOTABUG
Alias: None
Product: Fedora
Classification: Fedora
Component: kernel
Version: rawhide
Hardware: i686
OS: Linux
medium
medium
Target Milestone: ---
Assignee: Arjan van de Ven
QA Contact: Brian Brock
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2004-06-24 16:55 UTC by Jon Smirl
Modified: 2007-11-30 22:10 UTC (History)
0 users

Fixed In Version:
Clone Of:
Environment:
Last Closed: 2004-06-24 18:54:58 UTC
Type: ---
Embargoed:


Attachments (Terms of Use)

Description Jon Smirl 2004-06-24 16:55:53 UTC
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7)
Gecko/20040619 Firefox/0.9

Description of problem:
If I browse to www.dell.com everything works fine. If I then click on
the Small Business link and browse to
http://www1.us.dell.com/content/default.aspx?c=us&cs=04&l=en&s=bsd
net performace goes to zero. Another site with zero performance is
www.zmaxtech.com. Many others like slashdot, yahoo, linuxtoday work fine.

Backing off to kernel-smp-2.6.6-1.435 fixes everthing. Both the Linus
vanilla and Redhat kernels have this problem.

I captured packets to the site and don't see anything wrong. At the
problem sites one TCP packet is transmitted every half second or so.
No obvious protocol errors. Boot Windows or a 2.6.6 kernel and the
site works fine.

I don't think it is the e1000 driver since some sites work fine and
others don't on a 2.6.7 kernel which show this problem.

pings work fine to the problem sites.

Version-Release number of selected component (if applicable):
kernel-smp-2.6.7

How reproducible:
Always

Comment 1 David Miller 2004-06-24 18:54:58 UTC
It's a bug in Cisco routers when NAT is enabled.
It corrupts TCP SACK blocks in packets, which causes
lots of problems for the connection.

It happens when a non-zero window scale is used for the
TCP connection, which is what the newer Linux kernels are
doing by default.

It's not a Linux bug, and Cisco needs to own up to this problem
and provide a fix for their customers.


Comment 2 Jon Smirl 2004-06-24 19:36:02 UTC
Is there a simple work around while we wait on Cisco?


Note You need to log in before you can comment on or make changes to this bug.