[comp.unix.microport] WD1006 Controller on 386/ix

dirk@altger.UUCP (dirk) (07/25/88)

Hi!

Does anybody out there have any suggestion how to use
the WD1006 (RLL, 1:1) Controller under 386/ix or 386/V ?

I found out that Interleave 6 is the fastest one, but it is
slower than using a regular WD1003 with interleave 2.
So why does MS-DOS increase disktransfers by 3 when using
the 1:1 interleave but 386/ix slows down rapidly ?

cu,
dirk :-)

james@bigtex.uucp (James Van Artsdalen) (07/26/88)

IN article <865@altger.UUCP>, dirk@altger.UUCP (dirk) wrote:
> Does anybody out there have any suggestion how to use
> the WD1006 (RLL, 1:1) Controller under 386/ix or 386/V ?

The WD1006 also comes in a non-RLL ST506 configuration, and was/is
available with or without the floppy interface on board.

> I found out that Interleave 6 is the fastest one, but it is
> slower than using a regular WD1003 with interleave 2.

This makes little sense to me: this is a *1:1 interleave* controller:
you should use 1:1 interleave.  If the controller buffers by reading
an entire track at a time, formatting with 6:1 interleave could really
slow things down if the controller isn't smart enough to read the
sectors as they come (and I think the WD1006 might be this dumb - if
not, then your chosen interleave would be entirely irrelevant).

> So why does MS-DOS increase disktransfers by 3 when using
> the 1:1 interleave but 386/ix slows down rapidly ?

Well, in theory, a sufficiently fragmented file system with a
sufficiently fast drive *might* see this effect.  If memory serves me,
the rotational time of a most hard disks is 17ms.  If you read an
entire track, but really only wanted one sector, the wasted time is
8.5ms.  That's a lot of time for something like the CDC Wren III.

However, if a file system gets this badly fragmented, it should
probably be cleaned up for speed.  In particular, I re-load my
/usr/local directory regularly so that emacs and other binaries are
kept unfragmented for quick loading by my 1:1 controller (WD1007/WA2).
-- 
James R. Van Artsdalen   ...!ut-sally!utastro!bigtex!james   "Live Free or Die"
Home: 512-346-2444 Work: 328-0282; 110 Wild Basin Rd. Ste #230, Austin TX 78746

seeger@beach.cis.ufl.edu (Charles Seeger) (07/26/88)

Since the WD1006 has a full track cache, interleaves other than 1:1 don't
really affect the transfer rate off the disk, i.e. it takes only one disk
revolution to read the whole track.  Thereafter, interleaved reads should
progress at the same rate regardless of the interleave factor.  At that
point, reads should only be limited by system hardware and software.  Note
that the 1006 has 70 ns ram on board, so it's pretty fast.  Have you
disabled the cache?  I don't remember if that's even possible, not having
the docs here.

My experience using these controllers (WD1003, WD1006) has been with the
RLL versions under (ugh!) DOS.  Never tried anything other than 1:1 on the
1006 (in a 20 MHz '386), but the 1003 peaked at 3:1 (both the same '386
and a 10 MHz '286).  Keep in mind that the 1.5x increase in transfer
rate for RLL probably accounts for the discrepency in interleave factor
on the 1003 (your 2:1 vs. my 3:1), with the net transfer rate being
about the same.

Still, I don't understand the 6:1 interleave being optimum for your
1006.  How strong an effect was the interleave factor?

Chuck
(someday I'll create a neat ~/.signature)

det@hawkmoon.MN.ORG (Derek E. Terveer) (07/27/88)

In article <4796@bigtex.uucp>, james@bigtex.uucp (James Van Artsdalen) writes:
> However, if a file system gets this badly fragmented, it should
> probably be cleaned up for speed.  In particular, I re-load my
> /usr/local directory regularly so that emacs and other binaries are
> kept unfragmented for quick loading by my 1:1 controller (WD1007/WA2).

I don't quite understand this statement.  I would think that unless you are
creating and removing a lot of files in that directory (fs), in particular the
binaries you are talking about, you wouldn't be fragmenting.. this is, in
essence, a read/only situation.
-- 
Derek Terveer		det@hawkmoon.MN.ORG
			w(612)681-6986
			h(612)688-0667

jeffh@weycord.WEYCO.COM (07/31/88)

I've got a 1006 controller and don't understand the (other than 1:1)
problem. My 386-20MHz system loves it!!

chip@ateng.UUCP (Chip Salzenberg) (08/03/88)

According to james@bigtex.uucp (James Van Artsdalen):
>Well, I don't recompile emacs too often, but perl, dist (does anyone
>have dist working on SysVr3?) and the various news packages get
>recompiled all the time ...

WARNING:  MEMORY HOG ALERT:

Some of the more frequently-used programs can have their load times reduced
dramatically by setting the sticky bit (chmod +t).  This will load the
program text once and leave it in memory, waiting for its next invocation.
(As distributed, Xenix has the sticky bit set on /bin/ls.) I have found this
approach to be very effective with my text editor; Perl is another obvious
choice.

-- 
Chip Salzenberg                <chip@ateng.uu.net> or <uunet!ateng!chip>
A T Engineering                My employer may or may not agree with me.
        You make me wanna break the laws of time and space
                    You make me wanna eat pork

vandys@hpisoa1.HP.COM (Andrew Valencia) (08/04/88)

/ hpisoa1:comp.unix.microport / chip@ateng.UUCP (Chip Salzenberg) / 10:02 am  Aug  2, 1988 /
>Some of the more frequently-used programs can have their load times reduced
>dramatically by setting the sticky bit (chmod +t).  This will load the
>program text once and leave it in memory, waiting for its next invocation.

    I believe (unless they've REALLY changed things) that it'll only keep
it in the swap area.  It can still be swapped out and its memory resued.

					Andy