Formatting disk theory

[Chris Torek]

In article <619 at philmds.UUCP> hulsebos at philmds.UUCP (Rob Hulsebos) writes:
>I have the experience that most people who need to format a disk do not
>know anything about how this should be properly done.

Probably true.

>As I once read somewhere, [0xE5] is the [worst-case media] pattern for
>... FM-coding [aka `single density, which] ... is, however, something
>used in the past ....  now, most disks ... [use] the MFM-coding
>mechanism, also known as "double density". But the 0xE5 pattern is still
>used, which is not correct: double-density disks must be formatted with
>the 0xD6B6 pattern.

0xEC6D is listed as `media worst case' by the 4BSD Eagle formatter,
which expects MFM coding.  I have no idea whether this really *is* the
media worst case.

>QUESTION 1: Who knows more about this subject ? Any papers ?

Write to disk drive and media manufacturers; also look at standard
EDC/ECC references (Hamming codes).

>I also heard that either for VAX/VMS or Ultrix or BSD a special diskformat
>program exists which runs 40 or more passes over the disk, in an attempt
>to find all bad spots on it and also any marginal sectors.

`up to 48'.

>It seems that if you write the same data repeatedly to the same sector,
>and the sector is not 100% OK, after a certain time the sector will
>appear to be bad.

Rather, if you spend a lot of time on it, your chances of noticing that
it is bad are better.  Using different patterns is also wise, as it is
conceivable that somehow the media might get `permanently set' in some
bit pattern.  It is also a good idea to try to exercise the cabling,
and so forth: better to find faults *before* you put real data on the
disk.

>QUESTION 2: Is this program public domain?

No; it is under the usual Berkeley distribution clause.  It is also
not marked as being free of AT&T code, although I believe it must be.

>Has there ever been a proper description of the algorithm used ?

I am not sure what a `proper description' is, but it consists of
writing a set of test patterns, with ECC correction limited to a small
number of bits, possibly 0; this is easy on a Vax with Eagles, since
the ECC is done in software.  To make it go fast, format writes one
track at a time.  (Writing one cylinder at a time is not feasible since
the Massbuss limits transfers to 64KB.)

The test patterns are:

	/*
	 * Purdue/EE severe burnin patterns.
	 */
	unsigned short ppat[] = {
	0xf00f, 0xec6d, 0031463,0070707,0133333,0155555,0161616,0143434,
	0107070,0016161,0034343,0044444,0022222,0111111,0125252, 052525,
	0125252,0125252,0125252,0125252,0125252,0125252,0125252,0125252,
	#ifndef	SHORTPASS
	0125252,0125252,0125252,0125252,0125252,0125252,0125252,0125252,
	 052525, 052525, 052525, 052525, 052525, 052525, 052525, 052525,
	#endif
	 052525, 052525, 052525, 052525, 052525, 052525, 052525, 052525
	 };

I also have no idea what the worst case media pattern might be for
[2,7] RLL encoding.
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163)
Domain:	chris at mimsy.umd.edu	Path:	uunet!mimsy!chris