[lug] Best block size?

[Matthew Snelham]

On 17 Sep 2003 03:40 PM or thereabouts, John Karns wrote:
> > And of course, YMMV.
> 
> And indeed it does.  I've been using reiserfs on various servers for the
> past 18 - 24 months without any problems really worth mentioning ... until
> now.  I've run into a corruption issue with NFS (ver 3 compatibility
> option set) and 2.4.21 & 2.4.22 kernels with the HIGH_MEM (2GB) enabled
> w/ "HIGHMEM I/O support" / CONFIG_HIGHIO.  Looks like if I want to use
> more than 896 MB, I'll have to use another filesystem.

Interesting.  What kind of corruption have you seen, specifically?

I've used ReiserFS on a set of NFS servers for a client, with 1+ TB
fiber-channel volumes, HIGH_MEM (4 GB), HIGHMEM I/O, 2.4.19 kernel.org
sources, NFS v3 server, and v3 clients. (520+ actively mounting hosts)  And
the only issues have been with the NFS server code, and mounts randomly
going stale.  No good fix for that yet.  Though the problem doesn't show up
with the 2.5/2.6pre kernels... they're not suiteable for other reasons.  

I regularly use ReiserFS as a local filesystem on hosts with 2+ GB of RAM,
(including the machine I'm typing this on) so I doubt it's HIGHMEM or
HIGHIO alone that's the problem.
 
Very curious,
--Matthew
infinite at sigalrm.com

--
   "I speak Spanish to God, French to women, 
    English to men, and Japanese to my horse." 
      - Buckaroo Banzai