kparal at redhat.com
Thu Aug 12 08:42:34 UTC 2010
----- "He Rui" <rhe at redhat.com> wrote:
> I don't intend to cover all the memtest features, but think we should
> least check if there's a test result shown to reflect the test,
> the main goal to run a memtest.
Ok, I have waited for the whole test cycle to complete (10 minutes on my
machine). It really prints "Pass complete, no errors, press Esc to exit"
message. But - there is no report per se, the counter is just updated
(Pass +1). And it doesn't mean your RAM is ok. I have used memtest for
years and checked a lot of faulty hardware with it. From my experience
you need at least 2 hours of memtest checking to be quite reasonable sure
your RAM is ok. 4 - 8 hours are better. That's also the reason why it
continues over and over again.
Also, if some error is found, you don't have to wait to the end of the
test cycle. It is printed right away.
The message printed after one pass is complete is really not a test result.
It's just a message how to exit the test. The test results are updated
continuously - the screen is blue, you're good, the screen is red, you're
not good. The longer you wait (measured in hours) the more trust you have
in your hardware. So maybe that's the main misconception between us?
So, from my QA point of view:
1. We just need to check that memtest is available and it "works" (10
seconds passed or a single test finished is enough to check).
2. There's a very little probability that memtest would fail/crash on
some of its tests because of some bug - it will probably work ok or
won't work at all.
3. We don't want to check for RAM failures.
Of course I don't want to force my opinion to anyone. It's just my
personal idea how this test case could be improved.
More information about the test