Re: file placement and SAN

On Fri, 26 Apr 2002 15:15:47 -0600, Sean M <smckeownNO_at_BACKSIESearthlink.net> wrote:

>Michael Brown wrote:
>>
>> Thanks. I will say that your 'db file sequential read' averages about
>> 1/10 of what I had seen before.
>
>Maybe so, but the values aren't much different than the same database
>running in production on direct-attached disk. NAS, when pushed, will
>always be slower and look slower in tests, etc. But for real-world,
>end-user perception of "fast," we haven't seen that serious a
>degredation. When factored against all the other advantages of NAS, the
>choice is pretty easy for our needs.
>
>> One thing I forgot to ask was, what
>> is you db_file_multiblock_read_count (kind of important for
>> interpreting the scattered read result)?
>
>8.
>
>Thanks,
>Sean

Obviously Netapps had made some improvements since the last time I really dug into it. Your avg times are about twice what I see with my database using my newer disk arrays (SAN attached), and they are better than what I see on my older arrays. As far as the other advantages, that really is a personal preference issue. I prefer the access/tools I have with the SAN, and I know people who prefer the tools that are available with NAS.

One other thing that I am curious about is your CPU activity. When we evaluated a NetApps unit, our CPU usage went up about 25%. This raised two concerns with us, scability and Oracle licensing (this was in power units days). Do you notice a lot of system (as opposed to user) cpu utilization? We expected some since the TCP/IP stack for NFS is done in the CPU as opposed to the Fibre Channel activity which is done in the adapter card, but we were suprised at how much additional CPU was used.

Thanks,
Mike Received on Sat Apr 27 2002 - 16:01:40 CDT