Path: utzoo!attcan!uunet!mcvax!ukc!dcl-cs!aber-cs!pcg
From: pcg@aber-cs.UUCP (Piercarlo Grandi)
Newsgroups: comp.unix.microport
Subject: Re: How does Microport System V/AT handle bad blocks?
Summary: releases prior to 2.3 bungled horribly bad block handling
Message-ID: <452@aber-cs.UUCP>
Date: 21 Dec 88 00:44:03 GMT
Reply-To: pcg@cs.aber.ac.uk (Piercarlo Grandi)
Distribution: eunet,world
Organization: CS Dept., University College of Wales, Aberystwyth, UK
	(Disclaimer: my statements are purely personal)
Lines: 52

In article <326@focsys.UUCP> larry@focsys.UUCP (Larry Williamson) writes:

    In article <460@tarpit.UUCP> rd@tarpit.UUCP (Bob Thrush) writes:

	[ .... io errors on two drive system .... ]
    
    [ .... io errors as well .... ]
    
    We upgraded to 2.4 and errors have disappeared completely. We also replaced
    the disk, I couldn't bring myself to trust it.

The bad block handling code in 2.3 was horribly braindamaged. It did not
recover from soft errors, and then wrote random trash in random blocks.  The
disk instead you could have truested; it was clearly a case of environmental
(dis)adaptation of the format.

    I'm not sure why, but it seemed that the disk errors grew at an
    exponential rate.

A folksy description of a common problem follows.

Winchester disks are very delicate things. If operating temperature changes,
etc..., they suffer contraction/expansion of the surfaces, or of the heads
etc..., and what was previously recorded may become gibberish.  This does not
imply that the surface has become damaged though, simply that it has become
difficult to read back the recorded format.

The sumptoms are an increase of the number of soft errors, and then of hard
errors.  The cure is to reformat the disk. By the way, never trust a
preformatted disk; always reformat it on site, in the place where the machine
will be used, in its typical operating conditions.

    I would therefore suggest that you *very quickly*, get your 2.4 upgrade
    and install it.

The advantage of 2.4 is that bad block handling now is said to be ok.
previously if a read from a disk failed, it was not retried at all (even if
most errors are soft), and the buffer cache slot that was assigned to the
block to be read was not marked invalid. If and when written back to disk,
the previous contents of that slot would overwrite the contents of the disk
block, with astonishing results.

    I would also suggest that you verify your backups, you might be surprised
    by what is on (or not on) those tapes!

I would also suggest not to trust the current contents of your disks, unless
you check them. Note that I said *contents*, not just *structure*, i.e.  some
of your files contents may have been corrupted.
-- 
Piercarlo "Peter" Grandi			INET: pcg@cs.aber.ac.uk
Sw.Eng. Group, Dept. of Computer Science	UUCP: ...!mcvax!ukc!aber-cs!pcg
UCW, Penglais, Aberystwyth, WALES SY23 3BZ (UK)