Thanks for the insight Arnold.
Has anyone got an idea about the second part of the question.
> 2. Our average elapsed time for a scattered read is 26 Mb per seconds.
>
> I calculated this from de (level 8) wait trace as:
> (BLOCKS*blocksize/1024)/Sum(ELA/1,000,000)
> => (38,091*16/1024) / (22,780,281/1,000,000)
> = 26.13 Mb/second
> where:
> ELA = elapsed time for scattered reads (ela in microseconds) BLOCKS = p3
> for the scattered read event our blocksize is 16k and we
> multiblock_read_io_count=64 at a time.
>
> If this calculation is correct, it seems rather slow to me but I cannot
> prove this. Has anyone got some key figures or experience to back my gut
> feeling
Have fun,
Dikkie
------------------------------------------------------------------
>> Can't speak aboun Sun in particular as I'm used to EMC, but the
>> principles are the same. The striping is the important factor (over
>> mirroring), and in principle you should not have to worry about file
>> placement. In practice however, it may make a difference depending upon
>> how the stripes were created and across which physical devices. I've a
>> problem right now where two different applications use the same EMC
>> frame and we have IO issues. This is because the same physical devices
>> are used to make different logical devices, so as the applications
>> request disk from entirely separate filesystems, the physical disk is
>> experiencing thrashing.
>>
>> If your SAN experts build the logical devices correctly, you should
>> never have problems, but if not you will. They should have software
>> able to detect hot spots which they can then smooth out. I've found
>> that in practice, it's still best to put data files in one filesystem
>> and index files in another. This way there's less likihood of data and
>> index winding up on the same physical device.
>> Regards,
>> Arnold Hausmann
-----Original Message-----
From: Diederik_Dellaert
[Email address protected]
Sent: Tuesday, January 03, 2006 10:17 AM
To: LazyDBA Discussion
Subject: SAN Storage performance - data distribution and key figures
Has anyone got a configuration with oracle data warehouse sitting on Sun
SAN Storage I have two questions for you.
1. When using oracle on SANdisks, is it still usefull to distribute
datafiles over distinct SAN mountpoints or should SAN take care of
distributing data
2. Our average elapsed time for a scattered read is 26 Mb per seconds.
I calculated this from de (level 8) wait trace as:
(BLOCKS*blocksize/1024)/Sum(ELA/1,000,000)
=> (38,091*16/1024) / (22,780,281/1,000,000)
= 26.13 Mb/second
where:
ELA = elapsed time for scattered reads (ela in microseconds) BLOCKS = p3
for the scattered read event our blocksize is 16k and we
multiblock_read_io_count=64 at a time.
If this calculation is correct, it seems rather slow to me but I cannot
prove this. Has anyone got some key figures or experience to back my gut
feeling
Thanks for any feedback
Have fun in 2006.
Dikkie
--------
website: http://www.LazyDBA.com
Please don't reply to RTFM questions
Oracle documentation is here: http://tahiti.oracle.com
To unsubscribe: see http://www.lazydba.com/unsubscribe.html
To subscribe: see http://www.lazydba.com
By using this list you agree to these
terms:http://www.lazydba.com/legal.html
Oracle LazyDBA home page