| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 4063.1 | | WELSWS::FINNIS | | Fri Aug 31 1990 19:39 | 32 |
| And Here are my results..
From the CheapHD project with a RD53 and no real fastmem just A501
Any info to get the speed up in the hundreds would be appreciated.
DiskSpeed 2.0 - Copyright (c) 1989 by MKSoft Development
Device: dh3:
Comment 54Mb Partition No fastmem
Test Intensity: Med
8 Files/s Create
19 Files/s Open/Close
40 Files/s Scan
14 Files/s Delete
167 Seek/Read
Buffer Size 512 4096 32768 262144
--------- --------- --------- --------- ---------
Bytes/s Create 26294 25890 25663 26084
Bytes/s Create 27507 27464 27201 27264
Bytes/s Create 23002 27478 27217 27264
- Pete -
|
| 4063.2 | | POBOX::ANDREWS | I'm the NRA | Sat Sep 01 1990 13:41 | 6 |
| Any info to get the speed up in the hundreds would be appreciated.
THe best way is to play around with the interleave. I think mine is at
3 and I am getting figures that max out at ~125K with a spirit hda-506
and a 52.
|
| 4063.3 | try FIVE.... | PEEVAX::LAWSON | | Sun Sep 02 1990 18:05 | 13 |
| Try an Interleave of FIVE. I got transfer rates up to about 80Kb/sec.
Stan Gifford (peevax) was using another driver and was also peaking at
about 28Kb/sec. He too moved the interleave up to 5 and got the same
result.
WHilst 5 is the fastest, it seems as though SIX might be the most
reliable.
Keep us posted on your result.
Glenn
|
| 4063.4 | | PEEVAX::GIFFORD | My dunny was kicked down by chooks! | Sun Sep 02 1990 19:44 | 21 |
| Re .3
Yup on a RD52 with an Interleave of 5, 32k buffers gives me a transfer of about
60k - I am on the lookout for a rd53 to try now - it should be better because
of the faster seek time.
BTW. On this subject of hard disks, I have a a1000 (no comments please
I LIKE IT). I have to boot K/S then a workbench. The Workbench transfers controll
to the HD and executes DH0:S/Startup-sequence - The startup seems to cause the
disk to seek around alot. What I would like to try is to cluster all the startup
bits to the same cylinder, (along with the directory), so I get max speed.
Currently in spite of what I do, the startup takes always 1 min and 5 secs.
(I know this is quick, but I am a bit of a impatient so and so).
Anyone know about any 'placement' s/w (pref in public domain).
Also a de-frag for HD.
What are the thoughts re makeing a teeny tiny startup partition (about 1 1/2 mb)
Stan.
|
| 4063.5 | Six is more reliable ? | WELSWS::FINNIS | | Tue Sep 04 1990 16:14 | 21 |
| My Interleave for the RD53 is 3 at present..
Please explain ...'six is more reliable' did 5 give intermittent
problems ?
I am confused by the 32k Buffers ... ( Don't misunderstand I understand
the concept )... How does this relate to the command ADDBUFFERS or
is it specified in the mountlist.. Surely this is only applicable for
cacheing frequently read data and buffering writes.......
Wait a minute this is beginnining to sound like the Romans sketch in
Monty Python's Life of Brian....
How long did a format take you , anywhere near three and a
half hours from deep format to AmigaDos format/verify..
Also with the ST11R I could only specify 1021 Cylinders it
would not format if I specified any more...
- Pete -
|
| 4063.6 | Seems kind of slow to me | DECWET::DAVIS | You always get what you deserve | Tue Sep 04 1990 17:33 | 7 |
| The read times you are posting seem VERY SLOW. When I had a RD53
connected to my A500/supra controller, with interleave set
at 2 and step rate 1, the "large buffer" read rate was consistently
above 195K per sec. The fastest was ~220k reads/s. This is without
using the AmigaDOS "addbuffer" command.
mark
|
| 4063.7 | re: .5 and .6 | GIDDAY::LAWSON | | Tue Sep 04 1990 21:32 | 52 |
| re: .6
I'd say that the Supra controller is working in Interrupt/DMA mode. The
CHEAPHD project is basically a very simple signal converter which
allows the use of an XT type controller in "do it and keep testing to
see if its done" mode. The throughput is a function of how muchtime the
driver has to process info to/from the disk. I believe that it works on
a byte by byte basis.
re: .5
Firstly, my experience is with an RD52 (MFM) - Seagate ST11-M but I saw
exactly the same throughput maximums on 2 other controllers.
I had no percievable problems with an interleave of 5. Diskperfa told
me that this was the highest throughput. GUT FEEL made me feel that
there was a little more "clicking" occuring in the drive. Maybe some
retries were occuring.....
On a MFM type drive this suggests that the driver requires 5 sectors worth
of time to process the block's worth of data.
With an interleave of 4 it appears that the driver isnt ready to read/write
the data by the time the target sector arrives....so it waits for another
revolution.
with an interleave of 6, plenty of time, but 20% more disk revolutions
are necessary to process a multi-block transfer.
The speed improvement wasn't noticeable in "deep format mode" but was
definitely noticeable when doing Amigados format. I knew straight away
that something good was happening.
It took about 1/2 to 1/3 of the time (subjective!!)
To test out buffers, try a buffers = 1 and boot. it seeks like mad.
then try 32Kb worth of buffers (if you can afford it) and see the
difference.
Maybe things are different in an RLL environment. You might have to do
what I did...experiment. The deep format takes time (but not much -
comparitively). define Dh0: as only having say, 30 cylinders and then
format under DOS and then run Diskperfa. See where your transfer rate
peaks.
Hope that:
a. this helps and
b. I haven't said anything that is too wrong.
cheers,
Glenn
Oh yeah...old File system was very slow compared to FFS.
|
| 4063.8 | a change of pace | DECWET::DAVIS | You always get what you deserve | Thu Sep 06 1990 18:19 | 56 |
|
This is off the topic, but I thought I'd share some Diskspeed v3.1 results from
my new Supra 500XP host adapter and a conner cp3020(20Meg) and conner
cp3100(100meg) SCSI drives. These numbers were generated without any
additional buffers.
DiskSpeed 3.1 - Copyright (c) 1989,90 by MKSoft Development
Device: dh0:
Comment conner cp3020
Test Intensity: Med Performance Stress: None
10 Files/s Create
20 Files/s Open/Close
131 Files/s Scan
11 Files/s Delete
165 Seek/Read
Buffer Size 512 4096 32768 262144
--------- --------- --------- --------- ---------
Bytes/s Create 27497 118724 153129 178243
Bytes/s Write 29399 171196 288640 307662
Bytes/s Read 79309 195795 340798 363733
DiskSpeed 3.1 - Copyright (c) 1989,90 by MKSoft Development
Device: ih1:
Comment cp3100
Test Intensity: Med Performance Stress: None
11 Files/s Create
25 Files/s Open/Close
143 Files/s Scan
17 Files/s Delete
167 Seek/Read
Buffer Size 512 4096 32768 262144
--------- --------- --------- --------- ---------
Bytes/s Create 28873 145021 234851 251815
Bytes/s Write 29889 191261 397093 522819
Bytes/s Read 86508 304003 494356 546796
My system really "cooks" during I/O. Now to decrease my rendering times in
Silver...hmmmmm!
|
| 4063.9 | 6 or 9 for RLL .. Vote Now | WELSWS::FINNIS | | Fri Sep 07 1990 08:35 | 13 |
| Is it Controller Throughput or rotational latency here ?
ie With an interleave of 6 this is 1/3 of the circumference of the disk
in MFM (17 Sectors)
But with RLL it maybe 9 (again 1/3 of 27 ) [ 0-26 =27 ]
Once I've backed it up I will give it a try,
Did anyone buy the "Better Device Driver " from C'td quoted in the
original article ...
-= Pete =-
|
| 4063.10 | The answer is 9 | WELSWS::FINNIS | | Sun Oct 28 1990 18:06 | 39 |
| Well I finally got around to reformatting my drive.
I found that it was the Interleave of 9 for RLL that was required
and a buffer value of 64 buffers was needed.
Below are the new results
DiskSpeed 2.0 - Copyright (c) 1989 by MKSoft Development
Device: dh1:
Comment int of 9 and 60buff
Test Intensity: Med
9 Files/s Create
23 Files/s Open/Close
75 Files/s Scan
27 Files/s Delete
187 Seek/Read
Buffer Size 512 4096 32768 262144
--------- --------- --------- --------- ---------
Bytes/s Create 22282 59178 69814 72432
Bytes/s Create 22844 64527 80659 82316
Bytes/s Create 51894 64383 80829 82697
This is still without any real FASTRam so that may increase the
speed yet again.
I must admit that I did not expect to realistically tripple
the drive/controller performance by changing the interleave factor,
But I sure I am glad that I tried.
-Pete-
|