RE: I/O tuning... Allocating spindles to databases

Even if such access patterns are less than likely, it's still important to know how storage responds in worst case scenarios. Cache on the SAN/NAS is not a bad thing. I can't think of a situation where it would *degrade* performance merely by its presence. However, cache can mask architectural problems with the storage subsystem that cause real-world performance problems, therefore you have to evaluate uncached throughput *in addition to* cached throughput when comparing different storage solutions.

We've had a NetApp with rather fast disk and huge cache. Burst performance indeed can be good. But, the thing has effectively single gigabit fiber backbone and a single and heavily burdened storage processor that renders it incapable of sustaining more than maybe sixty (60) megabytes/sec sustained I/O out of a chassis that holds some 90+ 10krpm fiber disks. Cache does a great job of hiding that problem... Until the moment a single report overruns both Oracle's buffer cache as well as the NAS onboard cache. In seconds, performance is toast.

-----Original Message-----
From: Kevin Closson [mailto:kevinc_at_polyserve.com] Sent: Thursday, September 15, 2005 3:37 PM To: Mercadante, Thomas F (LABOR); jkstill_at_gmail.com; cmarquez_at_collegeboard.org
Cc: bob_murching_at_budco.com; oracle-l_at_freelists.org Subject: RE: I/O tuning... Allocating spindles to databases  

absolutely..it is VERY real life to have an access pattern that completely obliterates cache...

I can't count how many shops I've known
that completely glue a high end EMC to
the wall and lament the fact that there
is no way to disable the cache...

that was on particular models, I don't
claim to be an EMC expert by any means.

>>>-----Original Message-----
>>>From: Mercadante, Thomas F (LABOR)
>>>[mailto:Thomas.Mercadante_at_labor.state.ny.us]
>>>Sent: Thursday, September 15, 2005 12:33 PM
>>>To: kevinc_at_polyserve.com; jkstill_at_gmail.com;
>>>cmarquez_at_collegeboard.org
>>>Cc: bob_murching_at_budco.com; oracle-l_at_freelists.org
>>>Subject: RE: I/O tuning... Allocating spindles to databases
>>>
>>>But would something like this ever really happen in real life. We
>>>can all make hardware/software fail once we know what its weaknesses
>>>are.
>>>
>>>Hell, I can make a database server fail. Quite simple really. Walk
>>>into the server room and flip the breaker.
>>>Shuts down every time!
>>>
>>>-----Original Message-----
>>>From: oracle-l-bounce_at_freelists.org
>>>[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Kevin Closson
>>>Sent: Thursday, September 15, 2005 3:26 PM
>>>To: jkstill_at_gmail.com; cmarquez_at_collegeboard.org
>>>Cc: bob_murching_at_budco.com; oracle-l_at_freelists.org
>>>Subject: RE: I/O tuning... Allocating spindles to databases
>>>
>>> >>>
>>>>>>It isn't too difficult to write a simulation that will render the
>>>>>>cache useless.
>>>
>>>actually quite easy. create a file that is 100 fold larger than cache
>>>and do completely random 4KB transfers.
>>>
>>>smoked
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>>>>
>>>>>>
>>>>>>--
>>>>>>Jared Still
>>>>>>Certifiable Oracle DBA and Part Time Perl Evangelist
>>>>>>11+ years of trying to appear to know what I'm doing.
>>>>>>--
>>>>>>http://www.freelists.org/webpage/oracle-l
>>>>>>
>>>--
>>>http://www.freelists.org/webpage/oracle-l
>>>

--
http://www.freelists.org/webpage/oracle-l