[Veritas-bu] Performance test of tape drives
2004-12-14 09:39:41
Subject: |
[Veritas-bu] Performance test of tape drives |
From: |
thoke AT northpeak DOT org (Tim Hoke) |
Date: |
Tue, 14 Dec 2004 08:39:41 -0600 (CST) |
While Gregg has a good set of commands, that isn't a realistic test of the
devices. /dev/zero isn't really a good representation of data. That
being said, using /dev/zero will certainly give an idea of theoretic
throughput.
Building on Gregg's info, I think a better test for dd would be to use
/dev/random like this:
time dd if=/dev/random of=/dev/rmt/Xcbn bs=256k count=4000
Here is what I like to use to get an idea of what kind of performance to
expect:
/* Determine amount of data being used */
# du -sk /<Data_Path_With_Lots_Of_Data>
/* Mount a blank tape */
# tpreq -ev <some brand new tape id> -d <density> -p <pool name> -f
# /tmp/tape
/* Test a tar to tape with a block size of 64K (64K/512 = 128) */
/* Other factors to try along with there corresponding block size: */
/* Blocking factor = Block size */
/* 128 = 64K; 256 = 128K; 512 = 256K) */
# time tar -cvbf 128 /tmp/tape <Data_Path_With_Lots_Of_Data>
/* To test for a 256K Block (262144 in SIZE_DATA_BUFFERS) */
# time tar -cvbf 512 /tmp/tape <Data_Path_With_Lots_Of_Data>
/* Rewind and test again */
# mt -f /tmp/tape rew
# time tar -cvbf 128 /tmp/tape <Data_Path_With_Lots_Of_Data>
# mt -f /tmp/tape rew
# time tar -cvbf 128 /tmp/tape <Data_Path_With_Lots_Of_Data>
/* Unmount the tape */
# tpunmount -f /tmp/tape
/* Label the tape so it won't be frozen when selected for backup */
# bplabel -m <mediaid> -d <denslity -o -p <pool name>
HTH
-Tim
On Tue, 14 Dec 2004, MacKinnon, G. R. (Gregory) wrote:
> ida3248b AT post.cybercity DOT dk wrote:
>
> >Hello All
> >
> >I need to test our tape drives outside NBU to se what max throughput is.
> >
> >I thought there was a thread about this but I can't find it.
> >
> >I want to eliminate network/disk.
> >
> >We have a mix of Solaris, AIX, HP and Windows media servers, and we are
> >getting Linux media servers before long.
> >
> >Ideas/suggestions/howtos will be greatly appreciated.
> >
> >Regards
> >Michael
> >
> >--
> >Cybercity Webhosting (http://www.cybercity.dk)
> >
> >_______________________________________________
> >Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
> >http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
> >
> >
>
> For the *nix machines you can use the time and dd commands. Like so:
>
> time dd if=/dev/zero of=/dev/rmt/2 bs=256k count=4000
>
> Then multiply # of blocks (4000) by block size (256) which is 1,024,000
> and divide by # of seconds.
>
> Can't help much with the Wintel box.
>
> Gregg
>
>
>
>
> =================================================================
> Gregg MacKinnon Ford Motor Co.
> gmackinn AT ford DOT com 2101 Village Rd.
> Technical Computing Section Rm 1116, MD 1074
> (313) 594-3716 pager 7958343 Dearborn Michigan, 48124
> ==================================================================
>
>
>
> _______________________________________________
> Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>
|
|
|