Tags:
create new tag
, view all tags

Question

The primary problem is an increasing performance problem. Updates to topics are EXTREMELY slow. This includes attachments. As far as I can see it is not a memory leak, CPU cycles aren't being hogged by another program, and there are no processes on the web server waiting on I/O. The topics are stored on local disk, and it's a pretty fast disk. It has channel bonded ethernet interfaces, trunked at its switch, which has fiber GBICs trunked to the core switch.

I am primarily using one web for all topics, and the number of topics is relatively small (198).

It ran fine for about three months, and I didn't even have to restart httpd in that time. A configuration change (changing UserDir options) prompted me to restart httpd, and performance degraded inside of 24 hours to the point where attachments of 24K were taking

Viewing pages is not slow. Even with dozens of concurrent requests, pages are retrieved on a local net in under a second on average.

I've attached testenv output to this topic, setlib.cfg, and TWiki.cfg.

Environment

TWiki version: TWikiRelease01Feb2003
TWiki plugins: DefaultPlugin , SessionPlugin , ImmediateNotifyPlugin, CommentPlugin, SlideShowPlugin, SpreadSheetPlugin, TablePlugin
Server OS: Red Hat Linux 8, kernel 2.4.18-18.8.0
Web server: Apache/2.0.40
Perl version: 5.8.0
Client OS: Mixed: Windows & Linux
Web Browser: Mixed: IE6 & Mozilla latest/Firebird latest

-- IsaiahWeiner - 20 Nov 2003

Answer

Something similar was reported at SaveForever - the common factor appears to be Red Hat 8 or 9, but that's also used by other people without problems. Have you tried testing without any plugins enabled? It could be something is hooked in on Save.

Other than that, I can only suggest the following ideas:

  • doing a truss or strace or whatever on the TWiki process that's hung, to try to see what system calls happen around the hang.
  • putting tcpdump on the network interface might help determine if there's a DNS lookup or a remote HTTP INCLUDE going on.
  • reviewing all config changes or apps installed in the 3 months that Apache was up and running OK
  • checking for any NFS-mounted disks that are very slow

-- RichardDonkin - 27 Nov 2003

Well, I moved it to a machine doing nothing but Apache and TWiki now. The performance is identical, alas. The NFS mount bit you mentioned, Richard, has prompted me to turn off ypbind and autofs and amd, which I hope will terminate any eroneous tree-walking by the Plugins. I've tried modifying the Plugin load order, and that didn't seem to matter. What do you recommend as the best method to temporarily disable a particular plugin? I wasn't sure what the results of moving or renaming the files in lib/Twiki/Plugins would be. Thanks for your time.

-- IsaiahWeiner - 14 Jan 2004

It seems to be mostly resolved. Attachments don't take forever and neither do most topics. There is one scenario remaining for poor performance, but one of its influencing factors is the web browser being used. When that changes from mozilla-1.2.1-26 (as shipped by Red Hat) to a more recent revision, the remaining scenario is resolved.

-- IsaiahWeiner - 15 Jan 2004

Topic attachments
I Attachment Action Size Date Who Comment
elsecfg TWiki.cfg manage 20.9 K 2003-11-20 - 22:53 IsaiahWeiner TWiki.cfg
elsecfg setlib.cfg manage 1.4 K 2003-11-20 - 22:44 IsaiahWeiner setlib.cfg
htmlhtml testenv.html manage 10.0 K 2003-11-20 - 22:43 IsaiahWeiner Testenv output.
Topic revision: r4 - 2004-01-15 - 23:20:02 - IsaiahWeiner
 
TWIKI.NET
This site is powered by the TWiki collaboration platform
Ideas, requests, problems regarding TWiki? Send feedback
Copyright © 1999-2009 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.