Re: CPU Power or bigger Cache?

Oliver Kleis <ok*spam*news_at_nwb.de> wrote in message news:<3ECB3C5B.F26D4B34_at_nwb.de>...
> Hello NG,
>
> we will buy a new Oracle-Database-Server in the next week an so I have a
> question to the Oracle-Profis here at the NG ;)
>
> We have the choose between a dual P4 2,4GHZ Xeon with 512kb Cache or a
> dual P4 2GHZ with 2MB Cache. The price difference between the both
> System is ca 10000?.
> Now my bose want to know what`s the best for our company.
> I think that the System with the bigger Cache is faster but how much?
> Has anyone some benchmark between 512kb und 2MB Cached P4s or some
> experience with this?
> I have already search in the Internet (google) but I couldn`t find some
> usefull at all :-(
>
> I hope for help :)

I haven't seen any benchmarks posted for the Intel Xeon MP processors, but Intel did publish a paper describing effect of cache size on their P III Xeon CPUs some time ago. For a "typical" database workload, it was around a 15% increase in performance for a 2 MB cache, vs the 512KB cache size. Your mileage will vary. A far more significant issue is - whether or not Hyperthreading helps or hurts performance in your environment. To enable or disable 'SMT' - that is the question.

When you're performing maintenance (basically single user mode) - and not using partitioning, parallel query, I think that you're better off disabling processor multithreading with most OSes now on P IV Xeon MPs - if you have more than one physical CPU in the box. One thing that I have noticed, is that processor affinity seems to be working rather well, that a single thread is not distributed across all CPUs, but stays running on a single (logical or physical) processor.

I haven't benchmarked performance with and without hyperthreading enabled for Oracle 817, 920 on W2K Adv Server - but will be doing so shortly.

I read some time ago that the Linux 2.5 (dev) kernel knows the difference between a physical and logical CPU - but this won't matter for production usage until after the 2.6 kernel stabilizes.

btw- this is not just an Intel phenomena - others, including Sun - are looking to exploit this technology.

Pd Received on Wed May 21 2003 - 08:42:52 CDT