Bacula-users

Re: [Bacula-users] Bacula spool on SSD -- solid state drive performance testing?

2009-07-24 00:54:59
Subject: Re: [Bacula-users] Bacula spool on SSD -- solid state drive performance testing?
From: Marc Cousin <cousinmarc AT gmail DOT com>
To: bacula-users AT lists.sourceforge DOT net
Date: Fri, 24 Jul 2009 06:48:24 +0200
> In theory, the latency from random IO should be much closer to zero on a
> flash drive than on a thrashing hard drive, so I was hoping I might need
> only 1 or two 64GB or 128GB flash drives to provide decent spool size,
> perhaps not even raid-ed.
>
> In addition, SSD/flash drives should be silent and heat up the room less
> (although that latter effect will be small--10 watts vs 2 watts for each
> drive)

For spooling/despooling there should be no latency problems. You need 
throughput more than latency, and a standard hard drive will be as good as a 
SSD or even better if setup correctly.

All you really need is to be able to read and write big streams at the same 
time. So the real problem is to help your disk scheduler to be able to read 
while having a lot of data in the write cache.

We managed to do that by raising the read ahead on the disk array (with the 
blockdev command in linux). We have managed to get 300MB read and write at the 
same time with a disk array (I admit it costed a bit more than 2 intel SSD 
drives), but we have terabytes of spool capacity.

If you really want a SSD, I'd use it for the catalog's database if I were you. 
There, disk latency is often the main source of contention.

------------------------------------------------------------------------------
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users

<Prev in Thread] Current Thread [Next in Thread>