Bug 101734

Summary: kernel-smp-2.6.0-0.test kernels + ip_conntrack = slow xmit?
Product: [Retired] Red Hat Raw Hide Reporter: Thomas J. Baker <tjb>
Component: kernelAssignee: Arjan van de Ven <arjanv>
Status: CLOSED RAWHIDE QA Contact: Brian Brock <bbrock>
Severity: medium Docs Contact:
Priority: medium    
Version: 1.0CC: jmorris
Target Milestone: ---   
Target Release: ---   
Hardware: i686   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2004-05-12 07:00:47 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:

Description Thomas J. Baker 2003-08-06 13:20:06 UTC
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.4) Gecko/20030703
Epiphany/0.8.2

Description of problem:
Is it a known problem that on my e1000 interface, if iptables type
modules are insmodded, my outgoing traffic slows to a crawl?

Modules in question:

ip_nat_irc              5360  0
ip_nat_ftp              6000  0
iptable_nat            24996  4 ipt_MASQUERADE,ip_nat_irc,ip_nat_ftp
ip_conntrack_irc       72340  1 ip_nat_irc
ip_conntrack_ftp       73236  1 ip_nat_ftp
ip_conntrack           31788  6
ipt_MASQUERADE,ip_nat_irc,ip_nat_ftp,iptable_nat,ip_conntrack_irc,ip_conntrack_ftp
ip_tables              19712  2 ipt_MASQUERADE,iptable_nat

Without them, I get normal 100Mbs type transfers approaching 10MB/s but with
them installed, I'm getting closer to 50KB/s. If I remove the
modules, things work as expected. (I need though them so my ipaq can talk to the
rest of the world.) Or is this just an old software/new kernel problem?

I'm running 2.6.0-0.test2.1.29smp on a dual 3.0GHz Dell 650N.


Version-Release number of selected component (if applicable):
kernel-smp-2.6.0-0.test2.1.29

How reproducible:
Always

Steps to Reproduce:
1. run 2.6 test kernel
2. install iptables modules
3. transfer any file to another system
    

Actual Results:  Transfer speeds are generally below 50KB/s.

Expected Results:  Transfer speeds are in the normal 8-11MB/s.

Additional info:

I'm just trying to figure out of this is a software (iptables) issue or a kernel
issue. I can provide any other details that are needed.

Comment 1 James Morris 2003-08-06 15:15:20 UTC
Can you try loading each module separately and find out which one causes the
slowdown?

Also, do you have any iptables rules installed?


Comment 2 Thomas J. Baker 2003-08-06 20:04:38 UTC
ip_conntrack is what does it:

[tjb@wintermute incoming]# modprobe ip_conntrack
[tjb@wintermute incoming]# time scp radiohead.mov katratzi:/tmp
root@katratzi's password:
radiohead.mov                                   1%  836KB  49.1KB/s   19:50
ETAKilled by signal 2.
0.000u 0.007s 0:22.98 0.0%      0+0k 0+0io 224pf+0w
[tjb@wintermute incoming]# rmmod ip_conntrack
[tjb@wintermute incoming]# time scp radiohead.mov katratzi:/tmp
root@katratzi's password:
radiohead.mov                                 100%   57MB   8.3MB/s   00:06
1.408u 0.297s 0:12.08 13.9%     0+0k 0+0io 623pf+0w
[tjb@wintermute incoming]#


Comment 3 Thomas J. Baker 2003-08-21 19:01:32 UTC
This also happens with 2.6.0-0.test3.1.31smp.

Comment 4 Thomas J. Baker 2003-08-26 13:48:14 UTC
2.6.0-0.test4.1.32smp too.

Comment 5 Thomas J. Baker 2003-09-26 12:25:12 UTC
I just tried this on 2.6.0-0.test5.1.45smp and the problem is still there to
some degree. When transfering a 93MB file, the first 2.5MB or so is transfered
at about 50KB/s and then suddenly the transfer speed increases to about 8MB/s.
The total transfer time for a 93MB file averages about 1 minute with the
ip_conntrack installed and 11 seconds after removing it.

Comment 6 David Miller 2003-09-26 12:48:53 UTC
If it happens with 2.6.x, please report this
to the netfilter list (netfilter.org).  They play
with this stuff every day, so they're likely to figure out what the
problem is several orders of magnitude faster than I could.


Comment 7 Thomas J. Baker 2004-05-12 01:48:18 UTC
Fixed in recent kernels including kernel-smp-2.6.5-1.358.