List:Internals« Previous MessageNext Message »
From:Michael Widenius Date:March 5 2001 8:16pm
Subject:scsi vs ide performance on fsync's
View as plain text  
Hi!

>>>>> "Jeremy" == Jeremy Hansen <jeremy@stripped> writes:

Jeremy> We're doing some mysql benchmarking.  For some reason it seems that ide
Jeremy> drives are currently beating a scsi raid array and it seems to be related
Jeremy> to fsync's.  Bonnie stats show the scsi array to blow away ide as
Jeremy> expected, but mysql tests still have the idea beating on plain insert
Jeremy> speeds.  Can anyone explain how this is possible, or perhaps explain how
Jeremy> our testing may be flawed?

Jeremy> Here's the bonnie stats:

Jeremy> IDE Drive:

Jeremy> Version 1.00g       ------Sequential Output------ --Sequential Input-
> --Random-
Jeremy>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
Jeremy> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec
> %CP
Jeremy> jeremy         300M  9026  94 17524  12  8173   9  7269  83 23678   7 102.9  
> 0
Jeremy>                     ------Sequential Create------ --------Random
> Create--------
Jeremy>                     -Create-- --Read--- -Delete-- -Create-- --Read---
> -Delete--
Jeremy>               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec
> %CP
Jeremy>                  16   469  98  1476  98 16855  89   459  98  7132  99   688 
> 25


Jeremy> SCSI Array:

Jeremy> Version 1.00g       ------Sequential Output------ --Sequential Input-
> --Random-
Jeremy>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block--
> --Seeks--
Jeremy> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec
> %CP
Jeremy> orville        300M  8433 100 134143  99 127982  99  8016 100 374457  99 1583.4
>   6
Jeremy>                     ------Sequential Create------ --------Random
> Create--------
Jeremy>                     -Create-- --Read--- -Delete-- -Create-- --Read---
> -Delete--
Jeremy>               files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec
> %CP
Jeremy>                  16   503  13 +++++ +++   538  13   490  13 +++++ +++   428 
> 11

Jeremy> So...obviously from bonnie stats, the scsi array blows away the ide...but
Jeremy> using the attached c program, here's what we get for fsync stats using the
Jeremy> little c program I've attached:

Jeremy> IDE Drive:

Jeremy> jeremy:~# time ./xlog file.out fsync

Jeremy> real    0m1.850s
Jeremy> user    0m0.000s
Jeremy> sys     0m0.220s

Jeremy> SCSI Array:

Jeremy> [root@orville mysql_data]# time /root/xlog file.out fsync

Jeremy> real    0m23.586s
Jeremy> user    0m0.010s
Jeremy> sys     0m0.110s

<cut>

Couldn't the problem simply be that the SCSI array is caching things
in RAM before writing to disk while the IDE disk isn't but is flushing
things down to disk at once.

The behaveour of the above would be that when you do a lot of
read/write the SCSI would be much faster, as there is less read/writes
involved, but flush would be slower as there is always unflushed data
in the RAM.

Regards,
Monty
Thread