21 agosto 2008

Glossario informatico


RAID-5: configurazione ridondante di un gruppo di dischi rigidi. Il vantaggio principale di questa configurazione è che in caso di guasto, si scassano sempre due dischi per volta.

--

RAID-5: redundant configuration of an array of hard disks. Its main advantage lies in the fact that it guarantees that the hard disks will always fail two at a time.

7 commenti:

Uriel ha detto...

Spare disk: it seems useless, until the 2nd disk fails and you miss the first alarm.

Uriel

RL ha detto...

I second Uriel statement :-(

Eugenio Mastroviti ha detto...

@Uriel: it is useless: the second disk fails while the spare is still reconstructing (been there, done that, got the scars)

fbxxx ha detto...

During my 8 years as a software engineer at a supercomputing centre I can remember at least five double RAID-5 failures - two during reconstruction, another one with rather major consequences after an undetected failure - but only a couple of successful recoveries. It is likely that we weren't made aware of success stories as statistics suggests that disk failures should have occurred more frequently.

Running a parallel filesystem in production over fault-prone storage controllers was also great fun, though scientific computing is not too mission critical and can tolerate the occasional filesystem corruption (though this was not our case, some dedicated computing centres maintain that rerunning simulations costs less than archiving a second safety copy of the result data).

Uriel ha detto...

Running a parallel filesystem in production over fault-prone storage controllers was also great fun,
----

Aha. Geat job, GPFS. Oh, yes. I was wondering a 512 CPU 1350 IBM Cluster performing like my old Amiga 500, due of GPFS.

Oh, what a times, when i was doing HPC.....

Uriel

fbxxx ha detto...

The problematic disk array was attached to a very new and experimental machine configured with a Lustre filesystem, we experienced fewer software issue than we had anticipated, but more hardware problems than expected.

On the other hand, in the span of six years we had very little trouble with GPFS and performance was good.

Uriel ha detto...

On the other hand, in the span of six years we had very little trouble with GPFS and performance was good.
-----

Try solve 13th order secular equation, as used for chemical physicians, using huge matrixes. You'll go crazy with latencies.

Uriel