Linux Stress & Torture Testing, Burning In Systems With Open-Source

[phoronix]

Phoronix: Linux Stress & Torture Testing, Burning In Systems With Open-Source

There's a new way to pound your Linux/BSD systems very hard for burning them in, checking the system's reliability, and stressing them to the max...

http://www.phoronix.com/vr.php?view=MTg4MjY

[A Laggy Grunt]

But does it check the answer against a known correct answer?

The reason Prime95 is a good stress-tester is not just that it will crash a slightly too-overclocked CPU very quickly, but it will also know if this CPU is inserting single-bit errors into whatever it's doing. It will run computationally demanding routines on as many cores as you tell it to, and it will know if the computer makes an error doing this even if it doesn't cause it to overheat and crash.

[darkbasic]

No, it doesn't. If you want such a test there are plenty of them on linux. Just google for "linux cpu stress test".

[GreatEmerald]

Yea, this is a pretty funny feature, but I'll stick with prime95 for actual testing.

[Michael]

But does it check the answer against a known correct answer?

The reason Prime95 is a good stress-tester is not just that it will crash a slightly too-overclocked CPU very quickly, but it will also know if this CPU is inserting single-bit errors into whatever it's doing. It will run computationally demanding routines on as many cores as you tell it to, and it will know if the computer makes an error doing this even if it doesn't cause it to overheat and crash.

Short of crashing if there's stability issues or if the upstream program has built-in checking, no. The customer this was done for would run Unigine, a CPU benchmark, and a disk benchmark all at the same time for stressing their systems whenever making a change to their software stack in looking for stability issues. This is intended to meet those needs of just hitting the system hard, or if you want to check on the system's cooling, or other purposes mentioned, since like in the Prime95 instance it just bashes the CPU.

[duby229]

But does it check the answer against a known correct answer?

The reason Prime95 is a good stress-tester is not just that it will crash a slightly too-overclocked CPU very quickly, but it will also know if this CPU is inserting single-bit errors into whatever it's doing. It will run computationally demanding routines on as many cores as you tell it to, and it will know if the computer makes an error doing this even if it doesn't cause it to overheat and crash.

Which prime95 still does. It seems all this support does is allow you to start more than one stress test at the same time. The capabilities of the stress tests are a different matter that pts doesn't have any control over.

[Nobu]

I just use stress for stress-testing. Run it, hard-crash, up-vcore or drop-freq; run it again, kernel-panic, up-vcore/drop-freq; run it some more, stress thread gets killed/dies/otherwise stops unexpectedly, up-vcore/drop-freq; run it for good measure, hours later still stable, up-vcore/drop-freq one more time to be sure, then call it good...ish (run it one more time to be sure it didn't destabilize due to drop in freq/increase in voltage). Of course, prime95-like testing would be more ideal, but I haven't had any trouble with this methodology...yet.

[Michael]

Which prime95 still does. It seems all this support does is allow you to start more than one stress test at the same time. The capabilities of the stress tests are a different matter that pts doesn't have any control over.

Indeed. The explanation from this customer I implemented the feature for is that for each change to their operating system they run 200 hours of tests of workloads that are relevant to them and can trigger certain behavior, etc, and generally consists of Unigine and a few other tests all running concurrently.

[Raptor650]

I tested this new functionality TOTAL_LOOP_TIME. But i need to get through the command output result-file-to-csv. The entire command is

TOTAL_LOOP_TIME = 60 Phoronix-test-suite result-file-to-csv /tmp/result.csv stress-run c-ray steam

[Michael]

I tested this new functionality TOTAL_LOOP_TIME. But i need to get through the command output result-file-to-csv. The entire command is

TOTAL_LOOP_TIME = 60 Phoronix-test-suite result-file-to-csv /tmp/result.csv stress-run c-ray steam

Huh?

You should do it:

TOTAL_LOOP_TIME = 60 Phoronix-test-suite benchmark stress-run c-ray steam
phoronix-test-suite result-file-to-csv [whatever name you saved the results] > /tmp/result.csv