Re: stochastic fair queueing in the elevator [Re: [BENCHMARK] 2.4.20-ck3 / aa / rmap with contest]

[Valdis.Kletnieks@vt.edu]

[-- Attachment #1: Type: text/plain, Size: 1768 bytes --]

On Mon, 10 Feb 2003 06:10:08 +0100, Jakob Oestergaard said:

> In stock 2.4.20 the interaction is horrible - whatever was done there is
> not optimal.    A 'tar xf' on the client will neither load the network
> nor the server - it seems to be network latency bound (readahead not
> doing it's job - changing min-readahead and max-readahead on the client
> doesn't seem to make a difference). However, my desktop (running on the

This sounds like the traditional NFS suckage that has been there for decades.
The problem is that 'tar xf' ends up doing a *LOT* of NFS calls - a huge
stream of stat()/open()/chmod()/utime() calls.  On a local disk, most of
this gets accelerated by the in-core inode cache, but on an NFS mount, you're
looking at lots and lots of synchronous calls.

In 'man 5 exports':

       async  This option allows the NFS server to violate  the  NFS  protocol
              and  reply  to  requests before any changes made by that request
              have been committed to stable storage (e.g. disc drive).

              Using this option usually improves performance, but at the  cost
              that  an unclean server restart (i.e. a crash) can cause data to
              be lost or corrupted.

              In releases of nfs-utils upto and including 1.0.0,  this  option
              was  the  default.   In  this  and  future releases, sync is the
              default, and async must be explicit  requested  if  needed.   To
              help  make system adminstrators aware of this change, 'exportfs'
              will issue a warning if neither sync nor async is specified.

Does this address your NFS issue?
-- 
				Valdis Kletnieks
				Computer Systems Senior Engineer
				Virginia Tech


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]