Amanda-Users

Re: performance tuning, gzip

2008-07-25 15:09:40
Subject: Re: performance tuning, gzip
From: Jon LaBadie <jon AT jgcomp DOT com>
To: amanda-users AT amanda DOT org
Date: Fri, 25 Jul 2008 14:03:52 -0400
On Fri, Jul 25, 2008 at 01:18:40PM -0400, Brian Cuttler wrote:
> 
> We have a Solaris E250 amanda server backing up two T1000 servers,
> also Solaris, hosting Lotus Notes.
> 
> Over time, we decided on HW compression, runs where long but they
> completed pretty reliably at the same time every day.
> 
> We tried an experiment, since we hadn't really tried SW compression
> since we upgraded the client systems, we used SW-client compression
> and removed the HW compression. Runs jumped to 22+ hours, but we where
> not seeing the work area filled (the data was smaller, and it was taking
> longer to get to us).
> 
> So I increased the inparallel parameter, which of course ramped up
> the load on the clients even further.
> 
> The question of which version of Gzip to run arose, we had a fairly
> old version and there is a newer-Sun version available, just didn't
> know how version sensitive we where. I know version of gzip (which
> we use on some partitons on these clients) is very version specific.
> 
> Is there a list of tested/approved gzip versions ? I didn't see one
> but may not have dug deep enough.
> 
> Current gzip
> $ /usr/local/bin/gzip -V
> gzip 1.2.4 (18 Aug 93)
> 
> Proposed gzip
> $ /usr/bin/gzip -V
> gzip 1.3.5
> (2002-09-30)
> Copyright 2002 Free Software Foundation 
> 

One thing to check is whether you have specified "best" for your
compression.  Gzip allows you to select from 9 levels of compression,
trading cpu time (and wall time) for extra compression.  Amanda
allows you to select "fastest" (aka level 1), "best" (level 9) or
default which is level 6.

I just ran a quick test on an 11MB text only file.  Level 9 took
three times as long as level 1.  Yet level 1 gave 83% of the compression
of level 9.  I like default level 6 which took 1.8 times as long
as level 1 and gave 97% of the compression of level 9.

BTW I also ran bzip2 on the same file.  It did 60% better than gzip
level 9, but took nearly 22 times as long as gzip level 1.

-- 
Jon H. LaBadie                  jon AT jgcomp DOT com
 JG Computing
 12027 Creekbend Drive          (703) 787-0884
 Reston, VA  20194              (703) 787-0922 (fax)

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