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: How important is on-board cache for Oracle?

Re: How important is on-board cache for Oracle?

From: Willem Dekker <willem_at_serc.nl>
Date: Sun, 12 Dec 1999 10:21:29 +0100
Message-ID: <832953$f5u$1@newshost.accu.uu.nl>


The Onboard Cache, a.k.a level 2 cache, does not prevent any disk access being made.
You could view the hierachy like this
 Processor --- Level 1 cache ---- Level 2 cache ---- main memory (RAM)

The level 1 and 2 caches temporarely store parts of the main memory. They make the main memory faster from the standpoindt of the Processor. The actual Level 2 cache size you 'need' is hard to measure it will be used fully anyway. If a large L2 cache really gives a noticable (10% or more) speedup of your application, depends on how large the working set of your application is, and how fast the main memory is. It also depends on the cacheing strategy. In short it is hard to predict. The easiest is to check the benchmarks (www.tpc.org) and see what size of L2 cache is used in a simular setting as yours.
Since Oracle is a very big program (18MB on Linux 8.0.5), a big level 2 cache (1 MB or 2MB ) helps to store the most used part of the oracle executable in the cache.
Thus a large onboard cache does not prevent any disk access from happening, it makes the oracle process run faster. If you have optimised your I/O subsubsystem and your application in such a way that your application is mainly CPU bound, a larger L2 cache helps. A processor with a large L2 cache mostly demands a large premium compared to the standard size.

"Parker Sorenson" <sparker_at_averstar.com> wrote in message news:384BF07B.678285B9_at_averstar.com...
> In determining what type of processors to use in my server, I need to
> determine just how much actual use the on-board cache would get used
> while running a typical Oracle/Webserver software system - so I can
> determine how much cache I should get.
>
> There will be many, many very similar DB requests - but most going to
> seperate areas of the database (repeated request types - different
> data).
>
> Will the increased cache (say, like 1-2Mb on board) really still prevent
> a lot of disk accesses (and thus DB accesses)? How could I quantify
> this? I know that the caching is done at the processor instruction-set
> level of data, so how could I begin to assess how much cache could
> actually get used by various types of software systems?
>
> Any help is much appreciated.
>
> NOTE: all references I made to "cache" refer to the on-board cache
> closest to processor.
>
Received on Sun Dec 12 1999 - 03:21:29 CST

Original text of this message

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