Hi Daniel,

We were using a software RAID-5 on top of hardware RAID-5 across 3
4-disk volume groups. (1 LUN from each array volume group built the
software RAID-5). So we were able to lose 3 disks in a worst case
scenario.

It seems to me that neither RAID-1 or RAID-5 can lose more than one
disk without losing data, please correct me if I'm wrong.

Our data is 70% write/30% read, so the write latency is important. The
filesystem is ZFS.  Thanks again.

Best Regards,
Jason

On 12/4/06, Daniel da Veiga <[EMAIL PROTECTED]> wrote:
On 12/4/06, Jason J. W. Williams <[EMAIL PROTECTED]> wrote:
> Hi Daniel,
>
> Thank you very much for your help and advice. After some examination,
> we discovered a couple of things. It looks like our storage array
> layout was really bad for the IOPS MySQL was throwing at it, as a
> result the InnoDB transactions started to back-up under heavy load.
> Changing the array layout from RAID-5 to RAID-1 as well as moving the
> logs to their own spindles corrected the issue. Also, moving the
> InnoDB fsync log flushing interval from every commit to a 2 second
> interval helped dramatically.
>
> We found the storage was the problem by looking at SHOW INNODB STATUS
> while looking at the SCSI IOP latency.
>
> Does this sound reasonable to you?
>

Disk IO is one of innodb's bottleneck anyway, but I doubt this could
hurt performance as you suggested, making it unusable. You're the one
with access to the system, and thus the only one who can test it and
be sure ;) . Making a RAID 5 should increase read performance (if you
calculate the best segment size), but the write operations would be
not as fast as with a RAID 1, and you're risking data loss if more
than one of your disks go away. I never trade security for speed, and
if I were you I would check for another option. What's the most
frequent operation (read/write) on your tables?

Anyway, glad you solved your problem.

Just out of curiosity, what is your filesystem?

--
Daniel da Veiga
Computer Operator - RS - Brazil
-----BEGIN GEEK CODE BLOCK-----
Version: 3.1
GCM/IT/P/O d-? s:- a? C++$ UBLA++ P+ L++ E--- W+++$ N o+ K- w O M- V-
PS PE Y PGP- t+ 5 X+++ R+* tv b+ DI+++ D+ G+ e h+ r+ y++
------END GEEK CODE BLOCK------

--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:    http://lists.mysql.com/[EMAIL PROTECTED]



--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:    http://lists.mysql.com/[EMAIL PROTECTED]

Reply via email to