ADSM-L

Re: Large restores (again)

1999-05-01 06:45:51
Subject: Re: Large restores (again)
From: Fredrik Lundholm <exce7 AT CE.CHALMERS DOT SE>
Date: Sat, 1 May 1999 12:45:51 +0200
On Fri, Apr 30, 1999 at 03:05:45PM -0400, Richard Sims wrote:
> >Is there an issue with tons of itty-bitty files?
>
> Yes: ADSMv3 Small File Aggregation is supposed to help, but can be undone
> by TXNBytelimit/TXNGroupmax settings.


Hmm, I've observed the effect under Sun Solaris aswell, the problem IMHO is
not an ADSM transfer/buffer settings problem as much as (in the case of
Sun Solaris, and probably NT aswell)  synchronous meta-data updates
in the file-system. That is, for each file and directory you create there
is searching, reading and writing taking place on the disk to update
the directory structure, these operations are NOT cached so the ADSM
client must wait for the operations to actually take place before the next
file can be created/written. So for a lot of small files, you simlpy
cannot just fill caches and let the disk do it's job.
(ei you can't hide the inherent latencies that exists in disk architectures)

Short answer: The reported rate of restored files is pretty good!

/wfr/
Fredrik Lundholm
exce7 AT ce.chalmers DOT se
<Prev in Thread] Current Thread [Next in Thread>