[comp.sys.ibm.pc.hardware] How much faster is a 16 bit SCSI Card than the ST02?

dmoyer@oiscola.Columbia.NCR.COM (Dan Moyer) (05/29/91)

 
 
Hi,
 
A few questions for the collective network knowledge:
 
I currently have a 386sx with a Seagate ST-02 and ST296N SCSI disk.
 
With CORETEST and with Nortons SI benchmark, I get a reading of about
220-300 K bytes/sec transfer time, and a 28-31 ms access time.
 
These figures seem reasonable for a 8-bit, dumb scsi board and
this disk..
 
I am considering purchase of a 16 bit board.
 
I currently operate DOS and sometimes Windows 3.0 now.   Sometime
in the future (6-12 months), I'd like to be able to use OS/2 2.0 or
Unix.  
 
 
1)  What performance increase can I expect with a adapter card
    like the IN2000 or ADAPTEC 1542B, using the same ST296N disk.
    (I assume I'll need to do a low level format to get 1:1
     interleave).   If anyone outhere has done gone from a 8 bit
     card to a controller card like IN2000, WDFAST, or ADAPTEC, 
     how much better has your disk performance become?  I'd like to get
     an idea of this, before I purchase a new adapter card.
 
2)  Does anyone have recommendations of what board works, (or which
    ones don't work) with the ST296N disk?
 
3)  If I also upgrade with a new disk, 80-160 Mbyte range- any
    opinions, recommendations?   I understand that QUANTUM has on
    disk cache.  What affects have you seen with a disk with cache
    built in? 
 
4)  Any SCSI Cache controllers you'd recommend?  Again, what has been
    your experience in performance increase of a caching controller
    vs a non caching controller.  (And also a caching controller vs
    software caching like Hyperdisk or Smartdrive)
 
5)  In planning to goto OS/2 2.0 or Unix in the future, what should
    I avoid purchasing?   Should I change my idea of getting a SCSI
    adapter alltogether, and just get a faster IDE or ESDI drive, and
    keep the old card?
 
  I read this newsgroup frequently, so either post, or email any
  information, comments, or opinions.
 
  Thanks in advance,
  Dan Moyer
  dmoyer@oiscola.columbia.ncr.com
  803-739-7681
 
 
 

rkushner@sycom.UUCP (Ronald Kushner) (05/29/91)

In article <396@oiscola.Columbia.NCR.COM> dmoyer@oiscola.Columbia.NCR.COM (Dan Moyer) writes:
>Hi,
> 
>A few questions for the collective network knowledge:
> 
>I currently have a 386sx with a Seagate ST-02 and ST296N SCSI disk.
> 
>With CORETEST and with Nortons SI benchmark, I get a reading of about
>220-300 K bytes/sec transfer time, and a 28-31 ms access time.

I had 200-400K on the same drive with an high performance GVP controller for
the Amiga 2000. My Quantum was more around 800K or higher. The problem I had
with the 296N when I daisychained it to the Quantum was when I did heavy
access on the Quantum it would send the heads out of range on the 296N,
tried numerious controllers, and determined it was a problem in the 296N,
since the 296N did the same thing with an ST-157N-1 later on. Someone
suggest it was a Seagate ROM bug, but I never bothered to investigate, I
sold it off instead. Older 296's didn't have that problem.

>3)  If I also upgrade with a new disk, 80-160 Mbyte range- any
>    opinions, recommendations?   I understand that QUANTUM has on
>    disk cache.  What affects have you seen with a disk with cache
>    built in? 

I can't comment on the cache, but Quantum makes a damn good drive, makes the
Seagate look like the pig it is. It doesn't hurt to have it. Plus from
ClubMac you can get a bare 100 meg Qunatun SCSI drive for under $350. Just
gotta lie and tell them you got a "mac cx" or something like that. They
think that an SCSI drive can't work in anything but Mac. 

>5)  In planning to goto OS/2 2.0 or Unix in the future, what should
>    I avoid purchasing?   Should I change my idea of getting a SCSI
>    adapter alltogether, and just get a faster IDE or ESDI drive, and
>    keep the old card?

Avoid any non-DMA system. When you multi-task, your CPU will be busy running
the hard drive when it could be running a program. I don't like IDE drives,
too many incompatibilitys when you try to daisychain. We had a Maxtor and an
Quantum and they both refused to work with each other. After calling around,
it was like, well - buy another Quantum or buy another Maxtor. No one
offered solutions to getting them to work together, except to eliminate one 
of the drives. We wound up doing that. 

>  Thanks in advance,
>  Dan Moyer
>  dmoyer@oiscola.columbia.ncr.com
>  803-739-7681
> 
> 
> 

--
Ronald Kushner                          Life In Hell BBS  +1 (313) 939-6666
P.O. Box 353                               14400 USR HST V.42 & V.42bis 
Sterling Heights, MI  48311-0353              Complete Amiga Support   
UUCP: uunet!umich!vela!sycom!rkushner     (We are not satanic, just NUTS!)

wlsmith@valve.heart.rri.uwo.ca (Wayne L. Smith) (05/29/91)

In article <396@oiscola.Columbia.NCR.COM> dmoyer@oiscola.Columbia.NCR.COM (Dan Moyer) writes:
>
> 
>220-300 K bytes/sec transfer time, and a 28-31 ms access time.
>These figures seem reasonable for a 8-bit, dumb scsi board and
>this disk..
> 
>I am considering purchase of a 16 bit board.
> 
>I currently operate DOS and sometimes Windows 3.0 now.   Sometime
>in the future (6-12 months), I'd like to be able to use OS/2 2.0 or
>Unix.  
> 
I have had the same combo in my 286 (296N/ST02 controller)
I have low-level formatted the drive using the ST02 bios, using 1:1, 2:1, and
3:1 interleaves.  The one that worked best was 2:1.
At 2:1, both Spinrite and Core gave 425 +/- 25 KB/sec transfer rate.
The 1:1 gave less than 100 k/sec, and 3:1 was less than 2:1 (I forget what
it actually was).  I once heard that it was a fundamental problem with the
drive electronics that prevented good performance when formatted @ 1:1, but
I have talked to seagate and they say that it is the ST02 controller which
is limiting.  They say that the drive will do over 1M/sec with a 16 bit
controller.  I would say that, at least in a 286 system, the benefit/cost
in going to a 16 bit controller is nil.  When I tested the 296 head to head
with a ST251-1 (1:1 interleave, MFM) in the same system (using QAplus), 
both had identical transfer rates.  So, if you're satisfied with 17 sectors/
track 1:1 interleaved MFM, you'd be satisfied with 30 odd sectors/track
RLL at 2:1 interleave.

Someone else will have to tell you what to expect with a 16 bit SCSI
card, but I hear that Adaptec is best.

iverson@xstor.com (Tim Iverson) (05/31/91)

In article <396@oiscola.Columbia.NCR.COM> dmoyer@oiscola.Columbia.NCR.COM (Dan Moyer) writes:
>I currently have a 386sx with a Seagate ST-02 and ST296N SCSI disk.
>With CORETEST and with Nortons SI benchmark, I get a reading of about
>220-300 K bytes/sec transfer time, and a 28-31 ms access time.
> 
>I currently operate DOS and sometimes Windows 3.0 now.   Sometime
>in the future (6-12 months), I'd like to be able to use OS/2 2.0 or
>Unix.  
> 
>1)  What performance increase can I expect with a adapter card
>    like the IN2000 or ADAPTEC 1542B, using the same ST296N disk.

Under Windows 3.0, you will notice a (perhaps significant) decrease in
transfer rate if you go to a bus-mastering controller due to the double
buffering used to work around the physical-to-virtual translation.

Also, under DOS, none of the standard benchmarks work with the Always
card.  That is to say, they return results that are much higher than
reality.  This is because Always disables all interrupts during BIOS disk
transfers (thus the clock doesn't tick and transfers take zero time :-).
This trick can cause serious problems with serial port i/o.

>5)  In planning to goto OS/2 2.0 or Unix in the future, what should
>    I avoid purchasing?

Avoid non-bus-mastering host adapters.  Overhead under a multi-threaded OS
is very high for these critters (Always is not bus-mastering, Adaptec 154x
and WD7000 are).

>Should I change my idea of getting a SCSI
>    adapter alltogether, and just get a faster IDE or ESDI drive, and
>    keep the old card?

SCSI is easier to integrate and (with some drivers) more reliable.  ESDI is
faster in a single disk or single threaded system, SCSI is faster in a
mult-disk and multi-threaded system.  If you're going with Unix, you can add
tape and CD-rom to the same host adapter that drives your disk.

With most SCSI disks, if you develop a defect in an awkward place, you can
reformat to "remove" it; formatting maps out defective blocks, presenting a
flawless image of the media.  Some drivers, like the ones we sell with our
products :-), are smart enough to remap defects on-the-fly, so you'll never
even see one until you run out of spares.  Reformat and you can gain a new
set of spares.  With ESDI, you're SOL.

>  Dan Moyer
>  dmoyer@oiscola.columbia.ncr.com


- Tim Iverson
  iverson@xstor.com -/- uunet!xstor!iverson

bking@ersys.edmonton.ab.ca (Barry King) (05/31/91)

I have both an Adaptec 1540B and a DPT PM2001/95.  I now use the DPT 
board only.  I have it connected to a Conner CP3200F, 200 Mb SCSI-2 
drive.  In comparing the Adaptec to the DPT, I had some interesting 
results.  According to Core tests, for what they're worth, the DPT 
delivered faster seek times (18.1 msec for the DPT vs. 19.2 for the 
Adaptec).  I attribute this to the fact that the DPT uses a faster 
processor (MC68000, 10 MHz) and is a full 16-bit board meaning that all 
data paths are 16 bit.  This is not (I believe) the case for the 1540B - 
it uses and 8 bit processor.
 
As to throughput, the DPT was slightly faster but not significantly so.  
I would get about 785 Kb/sec for 64 Kb reads on the DPT, slightly lower 
for the Adpatec.  Again, this is probably due to the faster CPU on the 
DPT (therefore lower SCSI overhead perhaps).  I don't reallt trust Core 
for measuring the throughput.  I think something funny is going on 
because others with the same Conner drive have measured higher 
throughput.  The machine I'm using is a 386/33 ISA bus jobbie with an AMI 
BIOS.  So many things can affect disk I/O.  I've stopped worrying about 
the numbers.  When I reduce the blocking factor for Core I can push the 
throughput up past the 1.5 Mb/sec range.  I'm sure this is due to the 
fact that the Conner has a 64 Kb read-ahead cache on it.
 
Anyway, I think it's worth the 16 bit SCSI HBA for fast machines, 
especially since a 386/33 (and maybe even slower machines) spend a lot of 
time waiting for ISA bus activity to complete.  The Adaptec is a decent 
board - so is the DPT (I really like it).  I'm sure that there are other 
boards out there that perform equally as well.

Barry King              bking@ersys.edmonton.ab.ca
Edmonton Remote Systems:  Serving Northern Alberta since 1982

roy@cs.fau.edu (Roy Levow) (06/01/91)

In article <396@oiscola.Columbia.NCR.COM> dmoyer@oiscola.Columbia.NCR.COM (Dan Moyer) writes:
>I currently have a 386sx with a Seagate ST-02 and ST296N SCSI disk.
>
>With CORETEST and with Nortons SI benchmark, I get a reading of about
>220-300 K bytes/sec transfer time, and a 28-31 ms access time.
>  ... 
>I am considering purchase of a 16 bit board.
> 
>I currently operate DOS and sometimes Windows 3.0 now.   Sometime
>in the future (6-12 months), I'd like to be able to use OS/2 2.0 or
>Unix.  
> 
>1)  What performance increase can I expect with a adapter card
>    like the IN2000 or ADAPTEC 1542B, using the same ST296N disk.
>    (I assume I'll need to do a low level format to get 1:1
>     interleave).   If anyone outhere has done gone from a 8 bit
>     card to a controller card like IN2000, WDFAST, or ADAPTEC, 
>     how much better has your disk performance become?  I'd like to get
>     an idea of this, before I purchase a new adapter card.
> 
>2)  Does anyone have recommendations of what board works, (or which
>    ones don't work) with the ST296N disk?
>  ...
>5)  In planning to goto OS/2 2.0 or Unix in the future, what should
>    I avoid purchasing?   Should I change my idea of getting a SCSI
>    adapter alltogether, and just get a faster IDE or ESDI drive, and
>    keep the old card?

I recently replaced an ST02 on my 20MHz 386 with an Always IN-2000.  I
have a pair of ST296N's on the system.  Before upgrading, my trasnfer
rate was about 400+ KBps.  After the upgrade, with the same 2:1
interleave, there was no change in transfer rate.  But when I did the
low-level reformat to 1:1, the transfer rate doubled.  As an added
benefit, my floppy access time improved significantly.  (I didn't time
it before, so I can't quantify the increase, but it is quite
noticeable.)  By the way, the LL reformat was necessary to boot from
the hard disk with the new card even though I could access the disks
without it.

It is my understanding that the AHA-1542 would be a better choice for
a Unix system because of its DMA support; rather than the dual-port
RAM used on the IN-2000.  Someone else out in netland can probably
give you more details on this.  You also want to be _very_ careful
that the specific system you are planning to upgrade to supports the
controller you have purchased.  Since I have no plans to change from
the DOS/DESQview combo I am currently using, I didn't think it was
worth the extra $100 for the Adaptec card.

poffen@sj.ate.slb.com (Russ Poffenberger) (06/01/91)

In article <rkushner.4483@sycom.UUCP> rkushner@sycom.UUCP (Ronald Kushner) writes:
>In article <396@oiscola.Columbia.NCR.COM> dmoyer@oiscola.Columbia.NCR.COM (Dan Moyer) writes:
>>Hi,
>> 
>>A few questions for the collective network knowledge:
>> 
>>I currently have a 386sx with a Seagate ST-02 and ST296N SCSI disk.
>> 

[stuff deleted]

>>3)  If I also upgrade with a new disk, 80-160 Mbyte range- any
>>    opinions, recommendations?   I understand that QUANTUM has on
>>    disk cache.  What affects have you seen with a disk with cache
>>    built in? 
>
>I can't comment on the cache, but Quantum makes a damn good drive, makes the
>Seagate look like the pig it is. It doesn't hurt to have it. Plus from
>ClubMac you can get a bare 100 meg Qunatun SCSI drive for under $350. Just
>gotta lie and tell them you got a "mac cx" or something like that. They
>think that an SCSI drive can't work in anything but Mac. 
>

I was able to try both a Quantum ProDrive 105s and a CDC Wren IV on my Adaptec
AHA1542B. I was SORELY disappointed with the Quantum. The seeks were slow.
Benchmarks for both seek and transfer rate were dismal at best, and worst of
all, the numbers were very erratic. At times I would get reasonable rates, at
other times the seek times were horrible. I also know that this drive doesn't
perform any better on a Sun workstation. About the best transfer rate I could
get was 500Kbytes/sec. My Wren IV does 1.3Mb/sec. Even an old Wren IIIHH did
better than the Quantum. It could get 800Kb/sec.


Russ Poffenberger               DOMAIN: poffen@sj.ate.slb.com
Schlumberger Technologies       UUCP:   {uunet,decwrl,amdahl}!sjsca4!poffen
1601 Technology Drive		CIS:	72401,276
San Jose, Ca. 95110             (408)437-5254