Hard Disk Interleave
Clarence Dyson
Hi all, anybody know what the optimum interleave would be for a ST-412
hooked to a Shugart 1610-4 interface hooked to an Ampro LB. I finally
got the whole thing working again (bad chip on the 1610) and thought
it would be nice to set it up right.
Thanks,
Clarence
Roberto Waltman
cdy...@redshift.com (Clarence Dyson) wrote:
>Hi all, anybody know what the optimum interleave would be for a ST-412
>hooked to a Shugart 1610-4 interface hooked to an Ampro LB.
"Spinrite" would optimize old non-IDE disks in old IBM PCs to
anything between 3/1 to 5/1 interleave, depending on the disk and
processor speeds. I guess 5/1 would be a good starting point for
a CP/M system.
RW.
timo...@cyberramp.net
cdy...@redshift.com (Clarence Dyson) wrote:
>Hi all, anybody know what the optimum interleave would be for a ST-412
>hooked to a Shugart 1610-4 interface hooked to an Ampro LB. I finally
>got the whole thing working again (bad chip on the 1610) and thought
>it would be nice to set it up right.
I can't say for certain about that bd, but I use 8 to 1 on most Z80
based CP/M 2.2 systems.
Tim Olmstead
email : timo...@cyberramp.net
Visit the unofficial CP/M web site.
MAIN SITE AT : http://cdl.uta.edu/cpm
MIRROR AT : http://www.mathcs.emory.edu/~cfs/cpm
Clarence Dyson
timo...@cyberramp.net wrote:
>I can't say for certain about that bd, but I use 8 to 1 on most Z80
>based CP/M 2.2 systems.
That high. I think I used 3:1 on it. Might be why it's a tad slow.
Guess I'll just have to play and see.
Thanks,
Clarence
Clarence Dyson
rwal...@bellatlantic.net (Roberto Waltman) wrote:
>"Spinrite" would optimize old non-IDE disks in old IBM PCs to
>anything between 3/1 to 5/1 interleave, depending on the disk and
>processor speeds. I guess 5/1 would be a good starting point for
>a CP/M system.
>
I'm using 3:1 right now and it seems a tad slow. Guess I'll have to
try different interleaves and see which works the best.
Thanks,
Clarence
Axel Berger
*Roberto Waltman* wrote on Sat, 98-05-23 13:46 in comp.os.cpm:
RW>"Spinrite" would optimize old non-IDE disks in old IBM PCs
I don't think so. The optimum interleave is dependant not only of the
drive and controller but also the board. Thus spinrite won't
neccessarily find the optimum. But you can imitate it manually.
Starting with the highest interleave (17:1 for MFM) and testing the
spead with which the system reads, you decrease the interleave by steps
of one. The reading speed will get faster and faster until you have
gone beyond he optimum, when it will jump to a very low value. Of
course unless you find a way to just format very few tracks and test on
those, this will get extremely time consuming.
Tschö wa
Axel
timo...@cyberramp.net
cdy...@redshift.com (Clarence Dyson) wrote:
>timo...@cyberramp.net wrote:
There may be another factor to consider here : deblocking. What is the
sector soze on the drive? If it something other than 128 bytes, then
the way that the BIOS handles the deblocking could/will affect the
interleaving. If the physical sector size is greater than 128 bytes,
then you want no interleaving in the BIOS, and do t all in the track
format.
I have experiemented quite a bit with the interleaving of CP/M HD's
over the years as I always wrote my own BIOS's, and I found that 8 to
1 workes best, unless you have a VERY fast processor; say something
like a 10 MHZ Z80.
Don't be decieved by testing only with "time to load" tests. That is,
"How long does it take to start an app after I hit the <cr> key?" This
will give you a false reading. What you want is something that will
give you good OVERALL system performance. I like to use a battery of
tests to evaluate disk performance. Do some, or all, of teh following.
1. Use your favorite assembler to assemble the biggest program you
have.
2. Load Small C and have it compile itself.
3. Run FINDBAD on your HD.
4. If you do database type work then build a large (5MB) database and
re-index it.
Time all of these and record the results at each stage, then try the
next interleave. Make a grid of the results. This will make the
correct value easy to spot. It is the one that gives you the best
overall performance, doing real world work.
These are tried and rrue methods. I have been in the caching disk
controller business off and on since 1981.
Clarence Dyson
Thanks for the info Tim. I'll give them a try. Seems a very logical
way of doing it.
Clarence
David Director
Axel Berger wrote:
>
> *Roberto Waltman* wrote on Sat, 98-05-23 13:46 in comp.os.cpm:
> RW>"Spinrite" would optimize old non-IDE disks in old IBM PCs
>
> I don't think so. The optimum interleave is dependant not only of the
> drive and controller but also the board. Thus spinrite won't
> neccessarily find the optimum.
[snip]
Actually, Spinrite _will_ find the optimum, if you ask it to. It
does this by actually trying different interleave factors on one
cylinder, then testing the read speed. It will then recommend the
one that it calculates to be the optimum. I've rarely found it to
be wrong.
-- David
Peter Turnbull
David Director <ddir...@sungardams.com> wrote:
> Axel Berger wrote:
>>
>> *Roberto Waltman* wrote on Sat, 98-05-23 13:46 in comp.os.cpm:
>> RW>"Spinrite" would optimize old non-IDE disks in old IBM PCs
>>
>> I don't think so. The optimum interleave is dependant not only of the
>> drive and controller but also the board. Thus spinrite won't
>> neccessarily find the optimum.
> Actually, Spinrite _will_ find the optimum, if you ask it to. It
> does this by actually trying different interleave factors on one
> cylinder, then testing the read speed. It will then recommend the
> one that it calculates to be the optimum. I've rarely found it to
> be wrong.
Well, except that optimum for writing isn't always the same as the
optimum for reading, and I've seen at least one old slow system where
they were quite different. So Tim's answer was perhaps more approriate.
If Spinrite just tests reading, it's not the best way. Old PC AMI BIOS's
do both tests for that reason.
--
Pete Peter Turnbull
Dept. of Computer Science
[remove indy. from email address to reply] University of York
john r pierce
Axel_...@k2.maus.de (Axel Berger) wrote:
>*Roberto Waltman* wrote on Sat, 98-05-23 13:46 in comp.os.cpm:
>RW>"Spinrite" would optimize old non-IDE disks in old IBM PCs
>
>I don't think so. The optimum interleave is dependant not only of the
>drive and controller but also the board. Thus spinrite won't
>neccessarily find the optimum. But you can imitate it manually.
>Starting with the highest interleave (17:1 for MFM) and testing the
>spead with which the system reads, you decrease the interleave by steps
>of one. The reading speed will get faster and faster until you have
>gone beyond he optimum, when it will jump to a very low value. Of
>course unless you find a way to just format very few tracks and test on
>those, this will get extremely time consuming.
Maybe even easier... format the disk with no interleave at all, then read the
sectors in various orders... like 1,3,5,7,... to simulate interleave 2, etc.
I'd start at interleave 1 (none) and work up til I found the place where it
'jumps' to a faster speed. off hand, I'd bet its up around 7 (2 sectors per
turn max).
-jrp
----------------------------------------------------------------------
This posting has a invalid email address to discourage bulk emailers
Due to the ever increasing volumes of spam, I do not mix mail and news
----------------------------------------------------------------------
Axel Berger
*David Director* wrote on Tue, 98-05-26 10:13 in comp.os.cpm:
DD>Actually, Spinrite _will_ find the optimum, if you ask it to.
David, you have obviously missed my argument here. What you say is
quite right in itself, but to run Spinrite you have to plug the
controller into an IBM compatible board under MS-DOS. When you transfer
it back to your CP/M system afterwards, it is in no way garanteed, that
this interleave is still optimum. At the very least I would choose a
somewhat suboptimal interleave when doing it that way. An interleave
somwhat too big will slow the drive down by a couple of percent, one
only the teeny weeniest bit too small will do so by a factor of nearly
20, that is 1700 %.
Tschö wa
Axel
David Director
Absolutely right. I guess I shouldn't have jumped in after missing
the original posting; I hadn't realized that the drive was going to
be put back into a CP/M system. Thanks for the correction.
-- David
Clarence Dyson
David Director <ddir...@sungardams.com> wrote:
>Absolutely right. I guess I shouldn't have jumped in after missing
>the original posting; I hadn't realized that the drive was going to
>be put back into a CP/M system. Thanks for the correction.
>
>-- David
I guess I should have been clearer in the first message. Anyway, I did
find that with the 1610-4 Shugart and a ST-251 that 6:1 with a 50us
buffered step seem to be the fastest. Same drive in an IBM XT runs at
3:1
Thanks all,
Clarence