[Kolab-devel] Kontact Performance, once again...

[Chris Hoy Poy]

yes - this issue sucks, doesnt it! I can have a single user, running a
decent machine (dual-dual-core opteron64, 16GB of memory, on RAID-5) -
and the server running off something similiar, and I'll hit this issue. 

I've had users spit the dummy at me when Kontact locks up for 20-30
seconds while parsing folders :(

I think part of the problem is the lousy write performance on
synchronous NFS. It just has crap latency. the dimap is going to be
doing a lot of small i/o operations and I dont think NFS handles it very
well. 

Any latency issue there gets badly handled by Kontact - it has some
issues backgrounding the i/o.

This is a LOT better in KDEPIM-3.5.4, and I've had to drop the proko2
stuff in favour of just running vanilla Kontact (FedoraCore5, 3.5.4 if
anyone cares) Its got all the functionality we need - not sure on the
difference really. I preferred that to losing my job / health (those
machines weigh a lot when someone throws them at you!) 

The DIMAP functionality in Kontact is not fantastic. I think it lets the
team down. But again, its not until you run it over NFS that you see
these performance issues. And to anyone who wants to blame NFS here -
test Evolution in a similiar setup! ;) 

I have done some testing with NFS+cacheFS but havent had much luck -
havent tried too hard. Doesnt seem to help with the writes much. I had
more luck with Samba, but I'm loathe to introduce that just to assist
with this issue. It did seem to help a bit tho. 

//chris

On Tue, 2006-10-10 at 10:43 +0200, ITSEF Admin wrote:
> Hi all,
> 
> I'm not 100% sure whether this should go to the users list, but responses 
> there have been so few the last few times that I went for devel. Also, I 
> think this really needs the eyeballs of someone with insight in 
> Kolab/Kontact. :-/
> 
> I am *still* struggling with the performance of Kontact. I know there are 
> several issues in the bugtracker, but I can't quite place what I'm seeing in 
> any of them. One I found that might be related is 
> https://intevation.de/roundup/kolab/issue1390
> 
> The original problem we had was using Kontact together with NFS home 
> directories - to my great surprise, this seems to be an extremely unusal 
> setup, given the responses on kolab-user. In the original configuration, 
> Kontact would become very unresponsive as soon as a sync with the server 
> started. After some investigation, our first conclusion was that the poor 
> setup of our file server was part of the equation.
> 
> Since then, I have upgraded that server, so the home directories are served 
> from a RAID10 array. Some simple tests with large files indicate that this 
> array can easily saturate the 100Mb network connection our workstations 
> currently have. I have since moved ~/.kde/share/config and 
> ~/.kde/share/apps/kmail/dimap back onto the NFS share again (we had them on 
> the local drives as workaround - yes, *both* needed to be local to make 
> Kontact's performance bearable) for my own account to run some tests. And 
> again, when Kontact starts to sync, it becomes very sluggish as long as the 
> sync runs. Any scrolling becomes jumpy, even typing can be delayed. Switching 
> to a new folder becomes downright painful - and if any other network activity 
> starts (e.g. copy some large files from local drive to NFS share), Kontact 
> basically freezes.
> 
> When I monitor the network connection and the RAID array during the sync, I 
> can see that neither is maxed out during the sync. The client itself (proko2 
> after 2.1.4, currently revision 591884) runs on a P4/3.0GHz under SuSE 10.0. 
> CPU load isn't very high during the sync. I currently have one split 
> IMAP/DIMAP account (~230MB, ~70 folders) and two "pure" DIMAP accounts 
> (~100MB/32 folders and ~170MB/120 folders) in Kontact.
> 
> I admit, I'm still not 100% convinced that Kontact itself *is* the culprit, 
> but to investigate further, I'm certain I need input from folks with more 
> knowledge of what Kontact is actually doing that puts so much load on the NFS 
> system. I already had a look at strace and I can see many write actions 
> (kmailrc/kdeglobals) during a sync - plus the dimap folders themselves, of 
> course.
> 
> Any insight is welcome - and if any more testing on my part is needed, just 
> let me know.
> 
> Cheerio,
> 
> Thomas
> 
> _______________________________________________
> Kolab-devel mailing list
> Kolab-devel at kolab.org
> https://kolab.org/mailman/listinfo/kolab-devel
> 
> 
-- 
Chris Hoy Poy
System Administrator
DownUnder GeoSolutions
http://downundergeo.com.au