Recomendations for benchmarking software for Windows Servers (CPU/RAM/IO) please.

Forums › IT Pro and developers › Recomendations for benchmarking software for Windows Servers (CPU/RAM/IO) please.

new2nz

65 posts

Master Geek

Topic # 100266 6-Apr-2012 13:09

I've been tasked with collecting benchmarks from our clients servers so as to gain a measure as to how our software will perform on their systems.

We make BI software based on the MS SQL Server stack and it will use all the resources you can throw at it (Disk, CPU and Ram). The clients Server OS is Windows 2k8 R2

The first stage to get consistent bench mark of H/w performance and then later to collect stats based on the usage (which is client specific and we'll write our own tests for these).

I've found lots of comercial applications out there 3dMark, Passmark, Sandra (and we're happy to pay for them as needed) and a number of freebies such as IOMeter, IoZone, HDTune, Cerberus etc. But I'm not so sure if 'home pc utilities' are what I need - for example graphics on the servers is minimal to say the least :-)

I'd feel more comfortable if I had some recommendations from others who may have done this already please.

1 | 2

131 posts

Master Geek
+1 received by user: 8

Reply # 606098 6-Apr-2012 13:25

What's the clients server OS?

new2nz

65 posts

Master Geek

Reply # 606102 6-Apr-2012 13:36

The clients Server OS is Windows 2k8 R2 (have added that to the question)

tatbaird

131 posts

Master Geek
+1 received by user: 8

Reply # 606111 6-Apr-2012 14:09

Do you have full access to these servers? Are they DC's. member servers, file and print, etc?

Anyway, I just use the built in Task manager, Resource monitor, Event viewer and Reliability monitor.

Check out http://technet.microsoft.com/en-us/library/ee731897(v=ws.10).aspx

gzt

10519 posts

Uber Geek
+1 received by user: 1678

Reply # 606174 6-Apr-2012 17:13

You may have been tasked with that for a good reason, but the usual approach is to find the minimum acceptable performance level on an example configuration and specify a minimum hardware requirement to match.

Is there a reason for that approach? It will help in suggesting appropriate solutions.

new2nz

65 posts

Master Geek

Reply # 606208 6-Apr-2012 19:25

Thanks.

We do have a minimum requirement. However it's all data related and it's custom data and every client have varying requirements.

Also we (ahem) one client that insists that their subsystems as adequate and we're not so sure :-)

new2nz

65 posts

Master Geek

Reply # 606218 6-Apr-2012 19:26

I've had a look at PassMark Performance Test in detail and it does indeed look comprehensive....

Kyanar

3062 posts

Uber Geek
+1 received by user: 484

Trusted

Subscriber

Reply # 606228 6-Apr-2012 19:43

Yeah, PassMark is pretty good and comprehensive - it's one of the freebies included with my dedicated server so I've had the distinct joy of crushing a server into dust running a benchmark.

gzt

10519 posts

Uber Geek
+1 received by user: 1678

Reply # 606262 6-Apr-2012 21:37

Do the clients run other services and applications on the server in addition to yours?

new2nz

65 posts

Master Geek

Reply # 606284 6-Apr-2012 22:40

No we have dedicated servers.

insane

2293 posts

Uber Geek
+1 received by user: 376

Trusted

Subscriber

Reply # 606287 6-Apr-2012 22:40

As an aside I've found it quite interesting to see how in general software makers well over inflate the minimum system requirements for their applications. Out of the 100's of VM we host I can count on my fingers the number of servers which actually come close to using their allocated resources. The really hard worked servers are typically running very poorly optimized code.

If the servers are virtualised on a VMware based platform , VCenter Operations Manager can give you some pretty nice insight into the running of your server and let you know where a bottleneck may be.

Moving forward software vendors need to design horizontally salable apps, the idea of simply throwing more CPU, RAM and IO at a single server is old skool thinking.

gzt

10519 posts

Uber Geek
+1 received by user: 1678

Reply # 606325 7-Apr-2012 09:28

[edit: removed my comment here, i posted it to the wrong topic]

new2nz

65 posts

Master Geek

Reply # 606360 7-Apr-2012 11:41

Thanks All.

I've decided to trial Passmark for a while

Mark

1392 posts

Uber Geek
+1 received by user: 370

Reply # 606655 8-Apr-2012 13:22

This might be a way out there idea, but why not install your software onto the servers and see how well it runs ? :-)

Synthetic benchmarks are almost certainly not going to work the same way your app does, just have your system admin and the DBA (developer ?) who normally runs the software and have them tell you which they think is better. The DBA/Developer can tell you how well his app is going and the system admin can tell you how stressed the server was while doing it.

new2nz

65 posts

Master Geek

Reply # 606740 8-Apr-2012 16:59

In some cases it can take > 24hrs to load the data, and then there is overnight processing of about 1-x hours.

And in the past we would just do that, however we want to be able to give them indications on how particular hardware will likely perform.

Mark

1392 posts

Uber Geek
+1 received by user: 370

Reply # 606808 8-Apr-2012 19:34

Nothing like doing a job properly you know :-)

I can manipulate benchmarks to make practically any system shine! But the customers would see through it damn quick when they came to try it out in the real world :-)
Your best option for your reputation and happy customer factor is to do the testing for real, so what if it takes 2 days? Happy smiley customers not bad mouthing you is what you want....this is NZ every it guy knows every other it guy in their arena :-)

1 | 2