On Thu, 14 Sep 2006 01:58:38 +0000, Bob Jones wrote:
>
> "HansF" <Fuzzy.Greybeard_at_gmail.com> wrote in message
> news:pan.2006.09.14.01.17.58.831121_at_gmail.com...
>> On Thu, 14 Sep 2006 00:46:16 +0000, Bob Jones wrote:
>>>>
>>>> I often refer to the SpecMark info to get a general idea of the
>>>> capability
>>>> of a machine. Different SpecMarks for different machines - and similar
>>>> specmarks give me similar responses in load handling ability regardless
>>>> of
>>>> vendor.
>>>>
>>>
>>> Regardless of applications too? After all that is the most important
>>> variable out of the bunch.
>>
>> Absolutely, regardless of applications. As a first step.
>>
>>
>> The idea is to get a baseline for a known neutral application on one
>> machine. Pick an application that has recognizable characteristics that
>> may be applicable to the environment. Try to pick the application as
>> being a specific subset of the application, perhaps such as adding two
>> numbers.
>>
>> Do the same on a second machine.
>>
>>
>> The second step is to get an idea of how many times your application adds
>> two numbers. If it's a significant - and if the application is not
>> resource-limited - one can easily extrapolate the impact of switching to
>> the second machine, especially when holding all other factors constant.
>>
>> Iterate steps one and two with the next baseline application.
>>
>>
>> When using other people's benchmarks, Step 1 always is to get the
>> baselines. Step 2 is to see how relevant they are to your application and
>> factor that into the usefulness of the specific benchmark.
>>
>> Just because a benchmark doesn't exactly match a specific scenario doesn't
>> mean it's useless. It just means that a relevance adjustment has to be
>> factored in. But until that relevance is determined, the benchmark (either
>> formally or informally) can not be tossed. (IOW, it means one actually
>> have to think.)
>>
>>
>
> I could do that if time is an umlimited resource.
>
Yup.
Hence the need to take chunks bigger than adding two numbers.
Such as the documented load benchmarks created by others.
And approximate the relevance of that benchmark to the application.
--
Hans Forbrich (mailto: Fuzzy.GreyBeard_at_gmail.com)
*** Feel free to correct me when I'm wrong!
*** Top posting [replies] guarantees I won't respond.
Received on Wed Sep 13 2006 - 21:32:53 CDT
