Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: SATA drives for DB?

Re: SATA drives for DB?

From: Darren Dunham <ddunham_at_redwood.taos.com>
Date: Wed, 20 Oct 2004 22:56:07 GMT
Message-ID: <b2Cdd.33061$QJ3.1160@newssvr21.news.prodigy.com>


In comp.unix.solaris NetComrade <netcomradeNSPAM_at_bookexchange.net> wrote:
> In addition, do RAID controllers really give you that much more? We
> currently have about 10 spindles per DB, seems like any 'upgrade'
> would mean fewer spindles.

Different strokes for different folks. I've seen lots of cases where a stupid, slow raid controller with the NVRAM write cache gave a tremendous improvement because it improved write latency. For those jobs, that was all that was important. Not read speed, and not throughput.

Other cases, a well-tuned, set of high spindle jbods was the best they were going to do.

Is the 5200 maxed out? Is your application going to care if throughput were lowered? Many applications might not, because they're constrained elsewhere. Yours might.

> The maximum performance I saw on the A5200 was ~41MB/sec when attached
> to Sparc II and ~57MB/sec when attached to Sparc IV on _sequential_
> reads (creating 1G datafiles). I am sure on serial writes SATA
> solutions might be able to do as good on serial reads/writes, but will
> they perform well in db environment?

With an NVRAM front-end cache, I would expect write latency to improve. I've seen nothing that suggested a big difference in read latency, but I've never pushed one of those SATA arrays very hard. Just some very light demos.

> Our shop is OLTP by day, batch by night. IO is really not an issue
> during the day, and we don't have any historical data on IO
> performance, other then statspack logs (taken hourly every day)

> The best 'test' I could come up with, would be to attach a SATA arrays
> as a third mirror, and see if there is any performance degragation,
> but it would be nice to have some expectations, prior to buying an
> array.

I would expect that to be a good write test, but unless you set it to PREFER that array for reads, it might mask any slowness on the read side.

-- 
Darren Dunham                                           ddunham_at_taos.com
Senior Technical Consultant         TAOS            http://www.taos.com/
Got some Dr Pepper?                           San Francisco, CA bay area
         < This line left intentionally blank to confuse you. >
Received on Wed Oct 20 2004 - 17:56:07 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US