tpm

[mohan singh]

Hi ,

Can any body help me out to calculate the TPM for server , I have most of the N4000 server with diffrent cpu speed .
how TPM calculated , if I have diffrent cpu speed with diffrent no of cpu.

It will be a greate help to me.

Best regards
Mohan

[DCE]

You can check out the following link for industry standard TPC ratings:

http://www.tpc.org/

Dave

[Ted Buis]

TPM are dependent on the amount of memory, number of disks and I/O paths, as well as the CPUs. N4000 could support up to 8 CPUs, but the speeds ranged from 360 MHz (A3639A) to 750 MHz (A3639C). The TPC-C numbers that are published are not based on typical configurations, but are almost always done on maximum configurations, so your performance will always be substantially less. Also, your idea of a transaction is likely quite different from the way the TPC defines their transaction. If you have performance issues, I suggest you focus on seeing where your bottleneck lies, rather than gathering theoretical metrics that don't have any applicability to your environment.

[Tim D Fulford]

I agree with the above... but here is a rough breakdown..

N4000 4x 360MHz PA8500 24,200 tpm-c
N4000 8x 360MHz PA8500 43,400 tpm-c

N4000 4x 440MHz PA8500 27,500 tpm-c
N4000 8x 440MHz PA8500 49,300 tpm-c

N4000 4x 550MHz PA8600 33,500 tpm-c
N4000 8x 550MHz PA8600 60,000 tpm-c

rp7400 4x650MHz PA8700 44,000 tpm-c
rp7400 8x650MHz PA8700 71,000 tpm-c

rp7400 4x750MHz PA8700 48,000 tpm-c
rp7400 8x750MHz PA8700 80,000 tpm-c

I would use the above as a guide or RELATIVE performance, rather than absolute. so assume you went from 8x 440MHz CPUs to 8x 650MHz CPUs I would expect something like 44% (71,000/49,300) increase in performance/transactions.

Regards

Tim

[Ted Buis]

I have heard that in order to get 50,000 tpm, you would need a disk solution with enough disk so that you can sustain about 17,500 I/Os per second. So your numbers will often be limited by your storage solution if you are really trying to get to that number. IBM recently published a TPC-C that was quite large, but they used 2TB of RAM and over 6000 disks to do it. So some of the numbers aren't very appropriate for what we can expect to see in more typical configurations.

[Bill Hassell]

The TPM measurement is a proprietary series of tests, not a calculation based on CPU and disk speed. The reason is that decisions based on TPM need to reflect a somewhat real-world condition, a combintion of comptational elements along with I/O. As you might expect, a test that simply reads the disk one track at a time in sequential order does not reflect the somewhat random access nature of a busy database.

A common benchmark today comes from tpc.org and actually has several different tests with a different model for each test. Take a look at http://tpc.org/information/about/about.asp

[mohan singh]

Hi Every,

Thanx a ton for your valuable comments and suggestions . It looks that I will be able to reach up to some conclusion .


best regards
Mohan

[mohan singh]

Hi Tim and Ted,

Once again thanks, I have
One more query , As agreed upon one point that TPM is base on cpu clock speed, momory and disk i/o.
how much memory and disk i/o you considered while giving TPM detail in your pervious mail.

Best Regards
Mohan

[Ted Buis]

TPM depend on the database, operating system and many other factors. If you go to the TPC web site, www.tpc.org, you can look at all the published results for all vendors. One for the rp7400 which is the "C" version (A3639C) of the N-class shows 60,366.82 tpmC at $17.64/tpmC with 9 550MHz CPUs, 239 18GB 10K disks, 32GB RAM, Sybase, and HP-UX 11.0. Read the details of the Executive summary at:
http://www.tpc.org/results/individual_results/HP/tpcc.hp.rp7400.es.pdf or the full disclosure reports to get all the facts.

[mohan singh]

Hi All,

Thanx for your valuable tips.

Regards
Mohan