[NFS] NFS performance debuggins

[Adrian von Bidder <avbidder-xzBkAS4TQxQfv37vnLkPlQ@public.gmane.org>]

[-- Attachment #1.1: Type: text/plain, Size: 4149 bytes --]

Hi,

Environment:

several Debian based clients (Debian etch and etchnhalf kernels, this means 
2.6.18 or 2.6.24); Debian etch (2.6.18 kernel) NFS (v3) server.  Network 
seems basically ok ("ping -f -s 3000" works without losses, ifconfig and 
switch monitoring shows no errors) with no noticeable load.  Disks seem to 
have very little load either, NFS server has no other tasks.

Performance is sluggish :-(  Basically works, though -- no spurious errors.

tcpdump shows many "reply ERR 1448" etc. msgs whenever NFS activitiy is 
going on (both stat like with "find /home" or read/write with dd)

+++
16:49:24.778560 IP 10.0.1.2.2049 > 10.0.0.209.809066834: reply ERR 1448
16:49:24.790304 IP 10.0.1.2.2049 > 10.0.0.209.943279929: reply ERR 1448
16:49:24.801380 IP 10.0.1.2.2049 > 10.0.0.209.2001885801: reply ERR 1448
16:49:24.802173 IP 10.0.1.2.2049 > 10.0.0.209.860835666: reply ERR 1448
16:49:24.805286 IP 10.0.1.2.2049 > 10.0.0.209.1479697199: reply ERR 1332
16:49:24.807679 IP 10.0.1.2.2049 > 10.0.0.209.1096249460: reply ERR 1448
16:49:24.808358 IP 10.0.1.2.2049 > 10.0.0.209.2000902760: reply ERR 1332
16:49:24.809097 IP 10.0.1.2.2049 > 10.0.0.209.926298420: reply ERR 1448
16:49:24.809100 IP 10.0.1.2.2049 > 10.0.0.209.25105411: reply ERR 1332
16:49:24.817923 IP 10.0.1.2.2049 > 10.0.0.209.1366504235: reply ERR 1448
16:49:24.817927 IP 10.0.1.2.2049 > 10.0.0.209.352525071: reply ERR 1332
16:49:24.820397 IP 10.0.1.2.2049 > 10.0.0.209.269848846: reply ERR 1332
16:49:24.822097 IP 10.0.1.2.2049 > 10.0.0.209.1345540144: reply ERR 1448
16:49:24.822856 IP 10.0.1.2.2049 > 10.0.0.209.944780599: reply ERR 1448
16:49:24.825109 IP 10.0.1.2.2049 > 10.0.0.209.1395668559: reply ERR 1448
16:49:24.825112 IP 10.0.1.2.2049 > 10.0.0.209.1999335795: reply ERR 1332
16:49:24.827813 IP 10.0.1.2.2049 > 10.0.0.209.1685677906: reply ERR 1332
16:49:24.829439 IP 10.0.1.2.2049 > 10.0.0.209.1666084982: reply ERR 1448
16:49:24.829443 IP 10.0.1.2.2049 > 10.0.0.209.1415656037: reply ERR 1332
16:49:24.839013 IP 10.0.1.2.2049 > 10.0.0.209.911226680: reply ERR 1448
16:49:24.839017 IP 10.0.1.2.2049 > 10.0.0.209.1735414852: reply ERR 1332
16:49:24.841325 IP 10.0.1.2.2049 > 10.0.0.209.911358287: reply ERR 1332
16:49:24.842092 IP 10.0.1.2.2049 > 10.0.0.209.1364284211: reply ERR 1448
16:49:24.842800 IP 10.0.1.2.2049 > 10.0.0.209.258643250: reply ERR 1332
16:49:24.844256 IP 10.0.1.2.2049 > 10.0.0.209.1666017882: reply ERR 1448
16:49:24.844996 IP 10.0.1.2.2049 > 10.0.0.209.808595513: reply ERR 1448
16:49:24.845674 IP 10.0.1.2.2049 > 10.0.0.209.2000779112: reply ERR 1448
16:49:24.845677 IP 10.0.1.2.2049 > 10.0.0.209.1652175121: reply ERR 1332
16:49:24.847120 IP 10.0.1.2.2049 > 10.0.0.209.944722769: reply ERR 1448
16:49:24.847123 IP 10.0.1.2.2049 > 10.0.0.209.1682657874: reply ERR 1332
16:49:24.849334 IP 10.0.1.2.2049 > 10.0.0.209.944714835: reply ERR 1448
16:49:24.850873 IP 10.0.1.2.2049 > 10.0.0.209.1345861938: reply ERR 1448
16:49:24.918710 IP 10.0.1.2.2049 > 10.0.0.179.1936680564: reply ERR 1448
16:49:24.918719 IP 10.0.1.2.2049 > 10.0.0.179.1698508838: reply ERR 1448
16:49:24.921911 IP 10.0.1.2.2049 > 10.0.0.179.1633904741: reply ERR 1448
+++

Mount options: "rw,noatime,rsize=8192,wsize=8192,intr,hard,addr=10.0.1.2", 
it seems to pick tcp by default.  I had problems with UDP from some of the 
clients due to a strangely buggy VDSL switch in the path, so I haven't 
tried that again (I want to keep the DSL clients and the non-DSL clients 
identical if this is at all possible, so I can switch equipment around 
without reconfiguration.)

That performance is not optimal whith todays desktop environments (tons of 
small configuration files in both oo.org and kde) at login/program start on 
cold caches is one thing, but performance

Now where do I start debugging this?

-- 
Development costs of average proprietary and free software don't differ
radically because the methods are pretty much the same.  The huge
difference lies in the way the developers try to recoup their costs, not
in the costs they have to compensate.
        -- Florian Weimer on debian-security

[-- Attachment #1.2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 388 bytes --]

[-- Attachment #2: Type: text/plain, Size: 247 bytes --]

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://sourceforge.net/services/buy/index.php

[-- Attachment #3: Type: text/plain, Size: 362 bytes --]

_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs
_______________________________________________
Please note that nfs@lists.sourceforge.net is being discontinued.
Please subscribe to linux-nfs@vger.kernel.org instead.
    http://vger.kernel.org/vger-lists.html#linux-nfs