Method R Discussion Group

Method R helps developers, DBAs, and decision-makers optimize Oracle-based software in every phase of the software life cycle.

Discuss Method R products, techniques, and events.

Ask new questions or post your own responses.

Comparing two CPUs

Hi,

Which of them is the best way of comparing two CPIs

1.    Comparing them by their Clock rate?
2.    By its Spec rating (SPECint2000)?
3.    Please let me know if some other better approach.

Thanks,
Suraj Sharma
Suraj Sharma Send private email
Wednesday, March 5, 2008
 
 
Hi Suraj,

this depends a little bit on how your application uses the cpu. Our app does a lot of number crunching so I wrote a little pl/sql loop which calculates 2 million times "select 5*5 into  :var from dual" in order to get an estimate.

I found that - for our app - AMD Opteron 8216 was the fastest CPU while a current Sun SparcV9 was even slower then my old AMD Athlon XP 2200. The Opteron runs the loop in less the 30 secs while the Sparc needs 79 secs.

Regards
Friedrich
Friedrich Pfeiffer Send private email
Monday, March 10, 2008
 
 
Hi Friedrich,

Thanks a lot for your valuable input, but still I am not clear in the same. Suppose we want to move from one CPU to another. Let's say I want to move from Sun SparcV9 to AMD Opteron 8216.
Currentloy say our system is using 28 CPUs of Sun SparcV9, which is the best way to figureout how many CPUs of AMD Opteron 8216 equals to 28 CPUs of Sun SparcV9. My apology if I am still not clear on this.

Thanks,
Suraj Sharma
Suraj Sharma Send private email
Monday, March 10, 2008
 
 
Hi Suraj,

I apologize for my delayed reply due to some days off. Your request is very clear. With our (batch) application we can simply assume a (roughly) lossless scalability: If one typical application process completes within 30 sec. on an AMD Opteron we can run on an 8-cpu-machine 80 application processes in a 300 sec timeframe. With 8 Sparc cpus we could only run 30 of these processes. To achieve the same amount we would need 21 sparc cpus.

Of course there may be, depending on your application, a difference between the comparing one cpu to another on the one hand and comparing N cpus of type x with M cpus of type x on the other hand.

Perhaps your hardware vendor will lend you different machines so that you can do your on benchmarks.

One last point: our test procedure measures the performance of one cpu CORE (verified with sar –P –ALL 1 100) that equals a “oracle cpu”.

Regards,
Friedrich
Friedrich Pfeiffer Send private email
Tuesday, March 18, 2008
 
 

This topic is archived. No further replies will be accepted.

Other recent topics Other recent topics
 
Discussion Groups Main Page

Powered by FogBugz