Best chip for Oracle AMD, INTEL, SPARC
awatki...@googlemail.com
I know you are going to say that it all depends on the O/S, but if you
can forget that for the minute, does it matter what the CPU is when
your are compare the Intel Xeon Quads, AMD Quads and Sun SPARC chips.
Does the cache make a big difference, since you think then SPARC /
INTEL would be best and you would never consider AMD.
SPARC T2 = 4 MB L2
UltraSPARC IV = 2MB L2
INTEL Xeon = 2MB L2
AMD Opteron = 512K L2
Cheers
Andrew
Helma
Tons of books about Oracle performance, and this discussion about
different cache sizes is almost non-existent in this.
Seems that there are much bigger fish to catch when it comes to an
optimised system.
Having said that, i have seen a relation between hardware
specifications and oracle performance tuning. If the hardware is very
good, the performance of oracle is often very bad. This relation
( hey, let's call it Helma's Law ) is due to the fact that some
uninformed people tend to buy very good hardware if the performance of
the database is bad.
I've seen a few good running Oracle's on hardware that wasn't
topshelve. I haven't seen a good oracle DBA yet who claimed that the
hardware specifications needed to be improved.
Ok, my HELMA's LAW :
" The better the hardware, the lesser the optimalisation of the oracle
performance"
hmm. Perhaps a native speaker can improve this a bit. Is this correct
English anyway?
H.
bdbafh
Cache size is only one part of the overall architecture.
Perhaps a large cache is being used in order to try to make up for
higher latency (from memory)?
The use of an on-chip memory controller could help to reduce latency
as opposed to an off-chip memory controller.
How large is the penalty for misses in predictions?
How deep is the pipeline?
What is the front-side-bus speed?
What memory speed is supported?
Oh, and how is one to benchmark the CPUs if no OS is loaded?
Here is one such comparison:
http://www.anandtech.com/linux/showdoc.aspx?i=2158
hth.
-bdbafh
hpuxrac
We completed a bunch of benchmarking recently. It is really best to
do your own benchmarking with your own data and some custom SQL that
attempts to make call patterns that actually reflect what your company
needs to run.
But anyway, we tried to keep AMD and Sun in the running. The SUN
machines scale out well as # of connections go up but are very slow in
single threaded performance. Isn't each dedicated server session
essentially single threaded ... ( sorry of course they are ).
So for us the new Intel quad core Xeon's just absolutely tore up the
performance benchmarking we did. They have 24 meg of L2 cache I
think.
The AMD quad core's were not available at the time we did our
benchmarking.
Bob Jones
<awatki...@googlemail.com> wrote in message
news:0288f1f0-21c7-4c06...@k37g2000hsf.googlegroups.com...
Are they all the same price? Smaller cache can save you bigger cash.
John D Groenveld
awatki...@googlemail.com <awatki...@googlemail.com> wrote:
>I know you are going to say that it all depends on the O/S, but if you
>can forget that for the minute, does it matter what the CPU is when
>your are compare the Intel Xeon Quads, AMD Quads and Sun SPARC chips.
Have your system administrator order all three systems via try-n-buy.
<URL:http://www.sun.com/tryandbuy/>
Benchmark your application on each.
John
groe...@acm.org
joel garry
> <awatkins1...@googlemail.com> wrote in message
If you start talking about price, you need to watch Oracle licensing
very carefully. It is changing unpredictably. See
http://www.pythian.com/blogs/1009/recent-changes-to-oracle-se-licensing-rules-higher-price
for one example.
jg
--
@home.com is bogus.
http://www.signonsandiego.com/uniontrib/20080712/news_1b12milberg.html