-2

Just got some new Cores and RAM in last week, wamnted to stress test them and see if they were any good. Unfortunately I couldn't find any real resources on how to do this without an OD so I fired up StressLinux.

Dual QC Xeon 2.66's with HT, 16GB of RAM, fired up Stress with:

stress --cpu 10240 --io 256 --vm 16 --vm-bytes 1G --vm-keep --timeout 2000

Fires Back (amongst other things):

<-- worker 8129 got Signal 9
now reaping child worker process
failed run completed in 2009s

What does this mean? Where do I go from here?

[[ EDIT ]]

In addition, what are good and testing settings for such a machine? As of this edit I am using:

stress --cpu 30000 --io 2000 --vm 30 --vm-bytes 1G --vm-keep

And I still can't seem to get the CPU temps above <60C... Then again, it is a Dell PowerEdge 1900, and the airflow is insane... But shouldn't I be able to raise the temps a little more?

[[ EDIT AGAIN ]]

Maybe I should rephrase the question towards what I thought the obvious direction would be: How do I tax the processors to the extreme (what settings) to ensure I have rock solid chips (ie, before the second-hand warranty runs up).

I am using "heat" as a measure of taxation because it's the only way I know of. Low heat (only 20C above idle) would indicate to me that I am not pushing them as hard as I could.

A natural extension of that question may be: "Is there a way to tell just how much stress I am actually putting on the cores? Or how much more headroom I have?

Where's my oven?

Mike
  • 175
  • 1
  • 10
  • 3
    `But shouldn't I be able to raise the temps a little more?` You're clearly just not trying hard enough. To raise the temperature of the processors, remove the heatsinks and use the fans to suck air from your oven over the processors... all of which begs the question of why you're intentionally trying to cook your processors. – HopelessN00b Feb 21 '13 at 20:02
  • Touche. I have rephrased the question. – Mike Feb 21 '13 at 20:06
  • Buy spare parts. They're probably cheaper than spending your day on trying to heat up your server room via software. – Michael Hampton Feb 21 '13 at 20:20
  • Not spending a day, just a side thing. We're waiting on drives for an Array anyway, just figured I would see if I could max it out. Spares are fairly cheap. Like I said, just poking it. – Mike Feb 21 '13 at 20:26

1 Answers1

2

You're doing it WRONGEST.

Seriously - this is NOT how you "stress test" a system.
How hot the CPU gets is a poor measure of CPU workload -- I could have an lightly CPU easily exceed its designed operating temperature parameters (and probably trip its thermal protection circuit and shut off). I could have a CPU going full-bore at zero degrees if I were willing to cool it with liquid helium.
You are testing the efficiency of your cooling system, nothing else.

If you want to determine if a system is stable and functional simply put any general processor and memory intensive workload on it (try re-compiling an operating system: make world is still the standard stress test in the BSD community).

  • If the system crashes it's not stable, and you can start looking at components (RAM, CPU, Motherboard) to determine why.
  • If it survives you have one datapoint indicating possible stability, but remember:
    You can never tell how strong something IS, only how strong it WAS right before failure.

If you induce a failure in testing you know the component is now no longer serviceable. Whether it was bad when you got it or not you can't say - your stress test may have caused the failure (particularly if you're using "heat" as your arbiter of stress: Running a CPU above its maximum rated temperature is a great way to destroy it).

Modern computers sold by reputable vendors use high quality parts which are extensively tested by the manufacturer prior to sale. Companies like Intel rely on their good name to keep their business.
Accordingly, the chances of you having a bad CPU are relatively low. The chances of bad RAM, or a bad hard drive are significantly higher, but this is why we have ECC RAM and RAID arrays.

voretaq7
  • 79,879
  • 17
  • 130
  • 214
  • My question was evidently not constructive, which I fail to understand how a question can be such, but your question was very constructive. Thank you. I obviously did not understand the 'idea' of 'stress testing,' that fact, coupled with this answer (or any such answer, really) makes the combined whole patently 'constructive' for other users in the same situation, but I digress. Again, thank you. – Mike Feb 22 '13 at 14:26