Re: Oracle Benchmark Results for Different Hardware Configurations?

>>>>> http://www.tpc.org ???
>>>> Two issues with the site:
>>>>
>>>> 1. You may not find the exact configurations.
>>>
>>> In order for a benchmark to be published, the exact confirguration must
>>> be
>>> defined in the benchmark.
>>>
>> What I meant was OP's exact configurations.
>
> You'll never find the 'exact configuration' for the OP's system - even on
> the OP's site. 15 minutes after it's installed, it's probably changed.
>
>>> Whether you can purchase such a configuration is immaterial ... at
>>> least you know what is similar and what is significantly different.
>>>
>> We are not talking about whether you can purchase similar equipments. We
>> are talking about whether you can evaluate your configurations using
>> tpc.org as a guide or even a clue.
>
> Exactly ... and one can ...
>

How? What does a TPC benchmark tell you about the performance of your particular systems and applications?

>>>> 2. You will not likely archive the same performance with your
>>>> applications.
>>>
>>> Very true. Few people, or organizations, are willing to put in the
>>> effort to do that level of tuning - and the tuning only needs to be
>>> valid for a relatively short period of time.
>>>
>> Not just the level of tuning, but more importantly the type of
>> applications.
>
> Applications are what we tune.
>

Yes, but that's not the point.

>>> Yet - when trying to determine which hardware to get it is useful to
>>> have some starting point, no matter how artifically derived. There are
>>> a large number of variables in any benchmark and the TPC at least makes
>>> some attempt to define them and to hold some of them to a relative
>>> constant. (Then at least you have some ammo when trying to get a
>>> discount on the equipment management has already, behind your back,
>>> decided to purchase.)
>>>
>> If a large number of variables do not matter, then one could use a HP's
>> hardware benchmark to be the starting point of a Sun purchase.
>
> One can. Many do.
>

In that case, we could use anything as a starting point because even platform does not matter.

> At least TCP attempt to hold some of them (such as type & qty of workload,
> type of app) constant to give the analysis a bit of an even keel.
>

Again, how relevant is that to the real world, given both the apps and tuning processes are different? Received on Tue Sep 12 2006 - 23:02:45 CDT