Benchmark for HP-UX

[Ruediger Noack]

I need a simple benchmark program for HP-UX. The customer needs it to compare old K-class server with new demo L-class server before buying ...

Can anyone help me ?

Ruediger

[Sandor Horvath_2]

Hi !

Very very simple :
time primes 1 50000000 > /dev/null

but See the technical specifications

www.unixservers.hp.com

regards, Saa

[Pal Szabo_1]

Hi!

The simplest tools are:
a, top
b, uptime
c, sar
d, vmstat

HP offers also Glanceplus and Perfiview programs with their
systems, but these are
not a "simplest command line tool" category.

a, top is interactive, but useful, when you don't want to
produce graph from datas.
You can interactively compare the K-class and L-class load values.

b, Uptime shows your load values.It shows how many processes are waiting in processor queue.The measurement could be:

#!/bin/sh
while true
do
uptime > /tmp/kclasslog
sleep 5
done

You can easily process the collected data (kclasslog) with Excel, and you could produce fancy graphs.


c, Sar shows the various system parameters.
The data collecting is a bit complicated, but the man
explains it in detail.
The collected data is easy to process with Excel.
I suggest to compare the processor %idle times during the working hours.

d, vmstat show the various memory and processor parameters.The collected data is easy to process with Excel.

Regards:
Paul

[Cheryl Griffin]

Ruediger,
Contact your the local HP rep. They may be able to offer benchmark services allowing you to physically compare the two servers.

Also, you can get some benchmark performance info from here:
http://www.unixservers.hp.com/entrylevel/lclass/specifications/index.html
This site is for the Lclass machines.

The Servers home page is http://www.unixservers.hp.com/

[rick jones]

it really depends on what the customer wants to run... or if they want to run something at all.

Having said that, there might be some CPU95 results online for K's and L's at www.spec.org. Also, there might be some SPECweb96 results. Not sure if there are any SPEC SFS results. Sadly, the K is old enough now that any TPC results would have been long since withdrawn.

[Jim Welch]

Some places to start: (assuming you want to generate your own benchmarks rather than look up some existing measurements)

http://www.specbench.org/spec/
http://www.tpc.org/
http://www.rswsoftware.com/

[Christopher Caldwell]

Here's the poor man's unscientfic methodology you can use to determine relatively relative performance bechmarks:

#include
This mechanism doesn't consider I/O card configuration, applications, memory, yada yada, and it's loaded with assumptions -

HP's perfomance bechmarks are disguised on the website. They define the performance of one box relative to the performance of another box (mostly without regard to application or configuration). If you trace the history of the relations (theory of relativity), eventually you'll find something like

K580X6=17XD220

L2000X440X2=.6X17XD220

K580X6=1.67XL2000X440X2

So a K580 X 6 has roughly 1.6 X the OLTP capabilities
of an L2000X2.

An L2000X3 has roughly 1.1 X the OLTP capabilities
of an K580 X 6.

Your mileage may vary, but this example will show you want you want to know. Use the link to HP's hardware site to do the "math" for you make/model.

OLTP shouldn't be your only consideration, the L's are compelling for a number of reasons:
-lotsa horsepower
-small footprint (5 Ls per rack)
-large I/O (card count) capability
-lotsa hot swap stuff

I've never owned a K, but of all the HP hardware I have owned, HP did some of their best work in terms of maintenance and manageability on the L boxes.

[Ruediger Noack]

Hi Christopher,

I cant find this website ...

Ruediger

[Christopher Caldwell]

Go to
http://www.unixservers.hp.com/
then you pick your server (e.g. K)
http://www.unixservers.hp.com/midrange/kclass/index.html
then you pick specifications
http://www.unixservers.hp.com/midrange/kclass/specifications/index.html
then go to performance (at the bottom of specifications).
Here HP has posted "relative OLTP performance" which will allow you to compare boxes.

In the this example, the K is compared relatively to a D350.

At http://www.unixservers.hp.com/midrange/nclass/specifications/index.html
they compare HP relative OLTP performance as compared to K580 6-way.

At this point you can draw a direct corollary between K and N. If you ROUGHLY ASSUME Ls and Ns that are similarly outfitted (i.e. same number/type of processors) perform in a similar manner, then you can correlate L to K performance.