Re: [Evolution] Slow filtering under NFS



Try turning off indexing on your destination folders.  (select the
folder, goto file->folder->properties).

The indexing code in 1.0.x uses many small writes and reads and would
probably suffer *extreme* performance penalties because of nfs's
sychronous write semantics.  1.1.x's stuff should be a fair bit better,
since it uses bigger blocks, and most of the data is streamed rather
than written in random blocks.

I've seen this before with a cobol compiler, of all things.  It wrote
out code in 512 byte chunks, and compiling large libraries (in the order
of 10's of megabytes), it was very very slow via nfs.  Changed the
compilation process to write to /tmp (on solaris: tmpfs) and then just
cp the result back to nfs, and it changed the build time from about 8
hours to ~30 minutes.

The other operations - although we probably sync() more on folders than
we need to, shouldn't really suffer any significant impact vs anything
you'd expect from using nfs.


On Fri, 2002-05-31 at 10:44, Tad Truex wrote:
On Thu, 2002-05-30 at 21:06, Jeffrey Stedfast wrote:
Anyways, without a lot more information, it's impossible to say what
your problem is. It could be a slow network for all we know.


Sorry for the lack of info...  Here are more details,

1) Very small network (Server in the basement generally only one
satellite - the machine I read mail on)...  

2) The server is a 1.33MHz athlon, 1GB pc133, 5400rpm ata100 drive.

3) The network is all 100Mbit ethernet.

4) The client is an 800MHz celeron, 1GB pc133, 7200rpm ata100 disk


Both client and server are running rh7.2 and evo 1.0.5.  The mail is all
sitting in /var/spool/mail on the server (nfs mounted to the client)
having been fetchmailed from a remote pop server.  My home directory on
the client is an NFS (auto)mount from the server.  The filters I run
basically just sort mail in /var/spool/mail/tad into folders based on
which list they came from.  I tend to get a lot of mail 2-300 per day
and I check it a few times a day, so each filtering run generally
operates on ~50 - 100 messages.  I don't see any error or diagnostic
messages printed anywhere, so my guess is that things are working as
expected, just very slowly over the network.

Does that help (or have I as usual provided interesting but completely
tangential information...)

Thanks,
Tad


_______________________________________________
evolution maillist  -  evolution ximian com
http://lists.ximian.com/mailman/listinfo/evolution





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]