Messages in this thread | | | Subject | Re: Network slow-downs - someone follow this thread this time please! | From | cybertech@cybertec ... | Date | Mon, 28 Jun 1999 23:05:32 GMT |
| |
<br><font size=2 face="sans-serif">I'm reposting this information -- I posted it about a month back, but no one seemed to notice. I've solved the problem in the meanwhile the same way that svenden@pvv.org has -- cycling the interfaces. However, it's having to be done every 10 minutes, with a 5 second delay after downing -- HIGHLY annoying, even when done in cron.</font> <br> <br><font size=2 face="sans-serif">I've got 3 machines running 2.2.10. 2 of the machine are PPro200, 1 is PII 350. All are running 3COM 3C905/905b PCI controllers -- 5 of them in total, 2 running at 10mb, 3 at 100mb.</font> <br> <br><font size=2 face="sans-serif">After approximately 2 days of uptime, I will start to see ping times on the local lan jump to 7-20 seconds. As others have noticed, there is no loss -- just some damn high latency. This occurs whether i'm pinging another linux box, a windows box, or a cisco.</font> <br> <br><font size=2 face="sans-serif">It seems to be dependant upon the network load -- lighter loads lead to longer periods between problems. The problem ALSO is gradual -- it'll start at 4 second pings, then 7 second pings about 20 minutes later, than 30 minutes later it's up to 12-20 seconds. At this point the machine is almost unusable from remote. The problem doesn't seem to be triggered by data being forwarded -- I can route as much as i want thru the firewall (linux) w/o affecting it, but if the same machine is used for server activities it does begin to display the problem.</font> <br> <br><font size=2 face="sans-serif">My kernel is compiled w/gcc version egcs-2.91.66 19990314 (egcs-1.1.2 release)</font> <br> <br><font size=2 face="sans-serif">I normally compile the kernel w/ -mpentiumpro, but have gone back to -mpentium for the time being -- that did not change the problem at all.</font> <br> <br><font size=2 face="sans-serif">glibc is version 2.1 and 2.1.1, same symptoms under both.</font> <br> <br><font size=2 face="sans-serif">The network consists of a handful of windows boxes (98, NT4, Windows 2000), a cisco 2524, and the linux boxes. Nothing is dhcp enabled, ALL nics in all machine are 3com based, and i'm all out of relevant information. :)</font> <br> <br> <br><font size=2 face="sans-serif">David S. Stahl<br> President, CTCS/Dalsom Inc.<br> cybertech@cybertech.org</font> <br> <br> <br> <table width=100%> <tr valign=top> <td> <td><font size=2 face="sans-serif"><b>svendsen@pvv.org. (Øystein Svendsen)</b></font> <br><font size=2 face="sans-serif">Sent by: owner-linux-kernel@vger.rutgers.edu</font> <p><font size=2 face="sans-serif">06/28/1999 11:56 AM</font> <br> <td><font size=1 face="Arial"> </font> <br><font size=2 face="sans-serif"> To: linux-kernel@vger.rutgers.edu</font> <br><font size=2 face="sans-serif"> cc: (bcc: CyberTech/CTCS)</font> <br><font size=2 face="sans-serif"> Subject: Network slow-downs</font></table> <br> <br><font size=2><tt>Hello,<br> </tt></font> <br><font size=2><tt>I am experiencing quite drastically slow-downs on the network<br> performance on a dual P-II system running kernel 2.2.10.<br> </tt></font> <br><font size=2><tt>The machine currently has two network cards, one ISA NE2000 clone and<br> one Intel Ether Express 100 card, which is the card I have trouble<br> with. The machine works many as NFS-server.<br> </tt></font> <br><font size=2><tt>After upgrading to the 2.2 series we have from time to time<br> experienced severe slow-downs on the TCP performance, down to 1-10% of<br> the normal performance, measured by ttcp. According to /proc/net/dev,<br> we have no packet loss when the network slows down.<br> </tt></font> <br><font size=2><tt>When using ping from an Ultra 2 running Solaris 2.6, I record an<br> average ping response ranging from 100 to 4000 milliseconds, but no<br> loss, the ping response is just slow.<br> </tt></font> <br><font size=2><tt>The performance goes back to normal when I take down the interface and<br> reinsert the eepro100 module into the kernel. After I've done that,<br> the performance is fine for a couple of days or maybe weeks.<br> </tt></font> <br><font size=2><tt>I have tried to just take the interface down with ifconfig without<br> removing the module, and this seems to fix the problem, but only until<br> a large file is read through NFS. (I'm using the user level nfsd, but<br> I have had this problem using knfsd too.)<br> </tt></font> <br><font size=2><tt>When I measure the ping response while dd-ing a 60 MB file to<br> /dev/null through NFS on another machine, the response seems to be<br> fine until the whole file has been read, then the ping response grows<br> rapidly up to about an average of 100-200 ms, and after a hour or so<br> we end up with an average response of 3000-4000 ms.<br> </tt></font> <br><font size=2><tt>The traffic that goes through the NE2000-card seems to be unaffected<br> by what's happening on the eepro100 interface, the ping response is<br> stable way below 5 ms all time. I used to have an Accton 21140-based<br> card instead of the eepro100 card, and I experienced the same problems<br> using that card too.<br> </tt></font> <br><font size=2><tt>I have had seen this problem in 2.2.6, 2.2.7, 2.2.9 and 2.2.10.<br> </tt></font> <br><font size=2><tt>We have another box which is used as a workstation, but no SMP, and I<br> have on one occasion seen this problem there, and this box does not<br> have a SMP enabled kernel.<br> </tt></font> <br><font size=2><tt>I am not yet able to find a way to reproduce the problem, but I have<br> noticed that it somehow seems to be triggered by messing with the nfs<br> daemon (killing and restarting it, that is).<br> </tt></font> <br><font size=2><tt>I hope this is a problem that could be fixed, and I'll be happy to<br> provide more details, and otherwise do whatever is needed to find out<br> what's happening.<br> </tt></font> <br><font size=2><tt>--<br> Øystein Svendsen<br> </tt></font> <br> <br><font size=2><tt>-<br> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in<br> the body of a message to majordomo@vger.rutgers.edu<br> Please read the FAQ at http://www.tux.org/lkml/</tt></font> <br> <br>
| |