[nSLUG] Multiply-claimed blocks?

Daniel Morrison draker at gmail.com
Wed Aug 31 18:44:20 ADT 2011


On 31 August 2011 15:19, Dop Ganger <nslug at fop.ns.ca> wrote:
> On Wed, 31 Aug 2011, Daniel Morrison wrote:
>> I think that if a bit flip on disk once a month was "normal", our
>> computers would be crashing a whole lot more often.

I did say, though, "on disk".

> Not as often as you'd think; I have a firewall machine (AMD K6-350, which
> should say everything right there)

I have nothing against a K6 -- my favourite machine has, I think, a K6-3-400.

> with a stick of bad RAM in it and a hard
> drive that's reporting 0 hours lifetime remaining

You mean SMART says "this disk is in imminent danger"? It's not like a
fuel supply, when 0 litres left means you're dead. Estimates of
"remaining drive life" are pretty much meaningless. All we can say for
certain is that all drives will die eventually, some sooner than
others.

> chundering along to see
> how long before it will finally die. It never crashes, just causes stuff to
> segfault, and running monit allows any dead or hung daemons to be restarted.
> Current uptime is 28 days, since the last power outage that outlasted the
> UPS. Checking the logs it's had 31 errors related to RAM, or just over once
> a day. Since it started reporting errors about 4 years ago, it's only
> crashed and locked up once that I recall.

RAM errors that cause segfaults are different from disk bit errors.

In another email Mike Spencer said:

> In the usual course or operation, many files are written or re-written
> on new blocks at times invisible to the user. The current or canonical
> versions of some such files may be notionally "stable" of "static" but
> nevertheless reside in RAM at different times. I *think/infer* that
> this is true. If true, notionally static files could get silently
> munged in RAM and written invisibly to disk.

Linux will not write anything from memory to disk unless it absolutely
has to. libc, for example, will never get written to unless you
explicitly do so, e.g. package upgrade. Any file that does get written
to will have it's modification time updated. So you can try a sort by
time (ls -ltr) in various directories to see lots of files that have
not been written to since the O/S was installed.

Anyway, this is drifting way off topic. The original question was: Are
multiply-claimed blocks a common occurrence? A reliable sign of
impending disk failure?

My answers would  be no, and no.

The blog post that Ben linked in was absolutely fascinating, I
thought. Saved for future reference; thanks Ben!

-D.



More information about the nSLUG mailing list