Re: Fencing Cache in Disk Array for Oracle

The cache he is talking about is in the array and hence is not available to the SGA. I fear that you are in for some iteration. The amount of usable cache memory depends on the unique I/O signature of your database. With this particular array, and assuming you have 2Gb fibre you should expect reads and writes always down in single digit ms area. Tune with your choice of tool until you are as close to this as possible.

Allan

On 8/5/07, LS Cheng <exriscer_at_gmail.com> wrote:
>
> one of opinions here:
>
> http://blogs.smugmug.com/don/2007/04/27/the-perfect-db-storage-array/
>
> MySql focused but applies to any database vendor
>
> Thanks
>
> --
> LSC
>
>
>
> On 8/4/07, Binh Pham <binhpham15_at_hotmail.com> wrote:
> >
> > I'm currently working on a project that involves a new HP XP12000
> > Disk Array. The newly acquired array would have 240G of cache on the array
> > that allows data to be cached between the physical disks and the servers.
> > This model allows an amount of cache to be dedicated (fenced) to a specific
> > host, in this case, the production server therefore other non-production
> > related activities can pollute the production cache. The rest of the cache
> > can be shared among the rest of the non-production servers.
> >
> > My question is that how much cache should be allocated to ensure top
> > performance in terms of IO for the production server? I'm not comfortable
> > with throwing out a number and say that it is good. What I'm looking for is
> > some researchable facts that back up the number (for example, something that
> > I can mine from statspacks or similar that can be used to determine this
> > number).
> >
> > Thanks for any input.
> >
> > Binh Pham
> >
> >
> > ------------------------------
> > Messenger Café — open for fun 24/7. Hot games, cool activities served
> > daily. Visit now.
> > <http://cafemessenger.com/?ocid=TXT_TAGLM_AugWLtagline>
> >
>
>

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