Bacula-users

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

2009-07-23 13:35:22
Subject: Re: [Bacula-users] Bacula spool on SSD -- solid state drive performance testing?
From: Alan Brown <ajb2 AT mssl.ucl.ac DOT uk>
To: James Harper <james.harper AT bendigoit.com DOT au>
Date: Thu, 23 Jul 2009 18:29:25 +0100
James Harper wrote:

> I'd hazard a guess that for spooling, raw throughput is more important
> than random access seek time, unless you spool is fragmented.

You arte correct - IF you're only spooling one backup.

I may have anything up to 6 running simultaneously, some people on this 
list have much larger installations than mine.

As soon as you have more than one backup running there will be head seeks.

On a 4 spindle RAID-0 array, I found that aggregate throughput dropped 
more than 50% when spooling 2 100Gb files and unspooling 2 the same size.

Even just spooling one file while unspooling another had an aggregate 
performance hit of 25%.

On top of that the drives themselves tend to die quickly due to 
mechanical stress.

You won't see any of this if you're just spooling a single job, but in a 
largish installation with multiple tape drives and tens of terabytes of 
data, you _have_ to run simultaneous backups or you quite simply can't 
keep up.

AB




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