Troy Engel wrote:
>David Thompson wrote:
>> Peter =?iso-8859-1?q?Kjellstr=F6m?= wrote:
>>> raid5, #2 I can't get good read performance on two raid5s in a lvm stripe).
>>
>> Yeah, we've also seen performance issues with raid5 on the 9500 series
>> products. The benchmarks we run look good, but when we've put them into
>
>As per an old thread on the list, I have a new 9500S-12 and am seeing
>the same performance problems. But, I'm fairly sure it's because without
>the BBU installed write-cache is disabled. I have ordered the BBU and
>expect here soon, I'll see if I can't post some pre-BBU/write cache and
>post-BBU/write cache numbers, if I find a good test.
We throw caution to the wind and run write-cache-enabled without a BBU, and
still see a significant drop in performance compared to raid-1.
>Right now I'm running Arkeia on it as the backup-to-disk, and in the
>middle of the heaviest use I'm seeing about 50-60% I/O wait, horrible.
>I'm hoping the write cache is going to help this a lot. (*)
If you throw a sufficiently heavy (and multi-threaded) load at any disk array
of this type, eventually you can drive a system to I/O-wait. On our systems,
we look more at total disk throughput and characterization of workload
(sequential vs. random access), rather than just I/O wait.
Dave
Let me know what you find.
