[comp.unix.i386] Mylex SCSI Controller

palowoda@megatest.UUCP (Bob Palowoda) (09/20/89)

 Ok who is useing the Mylex DC376 SCSI Controller on a UNIX/XENIX machine.
 I'm interested to hear what the throughput is. The controller looks
 to be fairly new. This is what the ad says about it.

 * Has an onboard 80376.
 * The card uses the 32bit slot for 32bit IO
 * 13 Megabytes/sec transfer rate
 * Effective average .3ms seek
   (I hate when they say effective)
 * Emulates a WD controller (no special boot proms needed)
 * Software support for DOS, UNIX, Xenix and Nobell
 * Cache memory on board 1 or 4 meg simm modules.
   They didn't say total size but from the picture it looks like
   you can put 4 modules in.

 * Supports up to 7 SCSI drives, CD-roms, scanners etc etc.
 * They say the 32bit IO operates at 25 to 33Mhz.


 Seems like a hot controller. Might be on the expensive side.
 If anyone can please post what the real performance of the
 controller is.

 If anyone's serious about buying one of these
 Mylex Phone numbers:
 800-446-9539 outside ca.
 415-656-7857 inside ca.

 This is not meant to sound like an ad. I have nothing to do with
 Mylex or their products.

 I would like to see some benchmarks compareing the DTP ESDI 
 Cacheing controller vs the Mylex SCSI Cacheing controller.

 ---Bob



-- 
 Bob Palowoda    *Home of Fiver BBS*                   login: bbs               
 Work: {sun,decwrl,pyramid}!megatest!palowoda                           
 Home: {sun}ys2!fiver!palowoda   (A XBBS System)       2-lines   
 BBS:  (415)623-8809 2400/1200 (415)623-8806 1200/2400/9600/19200

larry@nstar.UUCP (Larry Snyder) (09/21/89)

>  Ok who is useing the Mylex DC376 SCSI Controller on a UNIX/XENIX machine.
>  I'm interested to hear what the throughput is. The controller looks
>  to be fairly new. This is what the ad says about it.

Hi Bob.  I have a friend in Indianapolis who just picked up one of the
new Mylex 33mhz motherboards with 4 megs and their 32 bit SCSI controller
to configure and run SCO 2.3.2.  After he gets this box up and running,
he will be tied into this (and other) newgroups and should be able 
to shed some light.



Does ESIX directly support the 16550A in their ASY drivers?  I wish
that SCO did.

-- 
Larry Snyder                                              SCO Xenix 2.3.2 '386
uucp: iuvax!ndcheg!ndmath!nstar!larry 
The Northern Star Usenet Distribution Site                    HST / PEP / V.22
Notre Dame, Indiana USA                            Home of the fighting Irish!

bote@csense.UUCP (John Boteler) (09/22/89)

From article <8191@megatest.UUCP>, by palowoda@megatest.UUCP (Bob Palowoda):
> 
>  Ok who is useing the Mylex DC376 SCSI Controller on a UNIX/XENIX machine.
>  I'm interested to hear what the throughput is. The controller looks
>  to be fairly new. This is what the ad says about it.
> 
>  * The card uses the 32bit slot for 32bit IO
>  * Emulates a WD controller (no special boot proms needed)
>  * Software support for DOS, UNIX, Xenix and Nobell
>  * Cache memory on board 1 or 4 meg simm modules.

Whose 32 bit slot? Intel's?  AMI's?  Mylex's new one?  Silicon Valley's?

The emulation is interesting...I assume they mean they emulate
ST-506 interface?

How Nobell of them to support UNIX!

Curious as to how a SCSI host controller can easily perform 
cache operations?


-- 
Bote
Old & Improved path!: uunet!comsea!csense!bote
New & Improved path!: {zardoz|uunet!tgate|cos!}ka3ovk!media!cyclops!csense!bote

palowoda@fiver.UUCP (Bob Palowoda) (09/23/89)

From article <111006@nstar.UUCP>, by larry@nstar.UUCP (Larry Snyder):
> 
> Hi Bob.  I have a friend in Indianapolis who just picked up one of the
> new Mylex 33mhz motherboards with 4 megs and their 32 bit SCSI controller

  I will also have access to one in a week or two. What I need is a good
benchmark to test it out. Does anyone know out in netland if such a 
piece of software exsists?

> Does ESIX directly support the 16550A in their ASY drivers?  I wish
> that SCO did.

  If what your asking do they enable the FIFO in thier ASY drivers, no. 
I don't think any of the UNIX suppliers or multi-port serial board
suppliers do. Of coarse thier is that asy driver you downloaded that
will enable it. I'm suprised they don't offer it.
I have had too many problems with ESIX's tty drivers. 
Actually the problems are nothing but setting up the correct 
conditions.

---Bob

-- 
Bob Palowoda  packbell!indetech!palowoda    *Home of Fiver BBS*  login: bbs
Home {sun|dasiy}!ys2!fiver!palowoda         (415)-623-8809 1200/2400
Work {sun|pyramid|decwrl}!megatest!palowoda (415)-623-8806 2400/9600/19200 TB
Voice: (415)-623-7495                        Public access UNIX XBBS   

larry@nstar.UUCP (Larry Snyder) (09/23/89)

> I don't think any of the UNIX suppliers or multi-port serial board
> suppliers do. Of coarse thier is that asy driver you downloaded that

The X5 kernel configuration kit that Interactive (in California -
Hollis claims that they have never heard of it) directly supports
the 16550A FIFO buffering.

-- 
                                                               Larry Snyder
                                      uucp: iuvax!ndcheg!ndmath!nstar!larry 
                                 The Northern Star Usenet Distribution Site 
                                                    Notre Dame, Indiana USA

nvk@ddsw1.MCS.COM (Norman Kohn) (09/24/89)

In article <8191@megatest.UUCP> palowoda@megatest.UUCP (Bob Palowoda) writes:
>
> Ok who is useing the Mylex DC376 SCSI Controller on a UNIX/XENIX machine.

I'd be a bit concerned: the board has an on-board cache, and
therefore can upset unix's notions of what order to do disk
writes in.  You might therefore incur a hard-disk inconsistency
unfixable by fsck if the system crashes in between writes.
I'd want to hear that the board had satisfied the driver writer
before I'd be comfortable bringing it up.

-- 
Norman Kohn   		| ...ddsw1!nvk
Chicago, Il.		| days/ans svc: (312) 650-6840
			| eves: (312) 373-0564

steve@nuchat.UUCP (Steve Nuchia) (09/25/89)

In article <653@fiver.UUCP> palowoda@fiver.UUCP (Bob Palowoda) writes:
>From article <111006@nstar.UUCP>, by larry@nstar.UUCP (Larry Snyder):
>> Does ESIX directly support the 16550A in their ASY drivers?  I wish
>> that SCO did.

>Actually the problems are nothing but setting up the correct 
>conditions.

No, there is more to it than turning on the FIFO's.  A driver that
was written without considering a FIFO will in general not try to
do enough work on each interrupt.  To take full advantage of the
550A's fifo you have to make different tradeoffs than you do for
8250's.

In particular, if you know you are likely to be working with 4-16
characters at a time you spend a lot more code setting up pointers
and such than you will if you expect 1, sometimes 2 characters.  Some
drivers will even return from the interrupt service routines before
completely filling/emptying the queues.

-- 
Steve Nuchia	      South Coast Computing Services
uunet!nuchat!steve    POB 890952  Houston, Texas  77289
(713) 964 2462	      Consultation & Systems, Support for PD Software.

palowoda@fiver.UUCP (Bob Palowoda) (09/25/89)

From article <111016@nstar.UUCP>, by larry@nstar.UUCP (Larry Snyder):
>> I don't think any of the UNIX suppliers or multi-port serial board
>> suppliers do. Of coarse thier is that asy driver you downloaded that
> 
> The X5 kernel configuration kit that Interactive (in California -
> Hollis claims that they have never heard of it) directly supports
> the 16550A FIFO buffering.

  I stand corrected, I poked around a bit and in asy.h thier is defines
for the fifo on the 16550. On ESIX it's enabled every 8 characters.
I was thinking about what you said about the comport not working 
properly under vpix at 38400. I seem to have problems anywhere > 9600.
I wonder if vpix uses the fifo on the 16550 or if thier is any dos
software that would use it to help soulve this problem.

---Bob

-- 
Bob Palowoda  packbell!indetech!palowoda    *Home of Fiver BBS*  login: bbs
Home {sun|dasiy}!ys2!fiver!palowoda         (415)-623-8809 1200/2400
Work {sun|pyramid|decwrl}!megatest!palowoda (415)-623-8806 2400/9600/19200 TB
Voice: (415)-623-7495                        Public access UNIX XBBS   

ken@gatech.edu (Ken Seefried III) (09/25/89)

In article <369@csense.UUCP> bote@csense.UUCP (John Boteler) writes:
>
>Whose 32 bit slot? Intel's?  AMI's?  Mylex's new one?  Silicon Valley's?
>

The card goes in the 32-bit slot on Mylex's MX (25&33) series motherboard.
According to Mylex, this slot is `based' on the Intel AT/32 32-bit slot,
but the board *will not* work in, say, an Intel 301 or 302.

>
>The emulation is interesting...I assume they mean they emulate
>ST-506 interface?
>

I believe the Mylex people said in emulated a WD controller until
(as yet undefined how) a driver is loaded into the board.

>
>How Nobell of them to support UNIX!
>

Ummmm...was this suppose to mean something, or just a stupid comment?

>
>Curious as to how a SCSI host controller can easily perform 
>cache operations?
>

Well...they have a microprocessor and a hoard of memory on board.
Think about it...or at least go review the recent discussion of
the DPT controller for some excellent commentary, pro and con, on
the issues involved in caching disk controllers...

	...ken seefried iii	
	   ken@gatech.edu

"I have lots of common sense...I just choose to ignore it" - Calvin&Hobbes

cliffhanger@cup.portal.com (Cliff C Heyer) (09/26/89)

Hopefully the Mylex will give 386 land the
*first* board that gives 900KB/sec disk I/O
like the Amiga (at Amiga prices hopefully)

davef@lakesys.UUCP (Dave Fenske) (09/26/89)

A couple of points on Mylex.  First, the company is a public company.  They
have been in business for a while.  The offer a two year warranty on all
their products (Compaq gives 90 days).

As for the question of the DC376 using the "AT32" bus being non-standard.  Well,
it is honestly no more non-standard than the memory cards in all of the 
different 386 boxes out there now, and frankly that doesn't seem to be much
cause for concern.

Mylex, I'm told, will be releasing EISA versions of their network card and the
disk controller later this year (when EISA becomes available).

Considering the dozens of computer manufacturers out there, I am rather pleased
that one, at least, has done something NOW, to address higher performance.

karl@ddsw1.MCS.COM (Karl Denninger) (09/27/89)

In article <22542@cup.portal.com> cliffhanger@cup.portal.com (Cliff C Heyer) writes:
>Hopefully the Mylex will give 386 land the
>*first* board that gives 900KB/sec disk I/O
>like the Amiga (at Amiga prices hopefully)

That's funny...

The WD1007/WA2 has been clocked here at 950KB/second (ESDI 10 Mhz).  I
understand that the WD1008 (15Mhz ESDI) has been clocked at over
1300KB/second, although I haven't seen that one myself.  The DPT boards 
have been clocked at that rate (950KB/sec) on cache misses.  On cache
hits the DPT board is off the top of scale on our tests; that is in 
excess of 4MB/sec!

While the Mylex board may very well be nice, it's a bummer that it's limited
to the Mylex motherboards....of course, given the differences in 32-bit
slots I understand the limitations...

--
Karl Denninger (karl@ddsw1.MCS.COM, <well-connected>!ddsw1!karl)
Public Access Data Line: [+1 312 566-8911], Voice: [+1 312 566-8910]
Macro Computer Solutions, Inc.		"Quality Solutions at a Fair Price"

bote@csense.UUCP (John Boteler) (09/28/89)

From article <19245@gatech.edu>, by ken@gatech.edu (Ken Seefried III):
> In article <369@csense.UUCP> bote@csense.UUCP (John Boteler) writes:
>>Curious as to how a SCSI host controller can easily perform 
>>cache operations?
> Well...they have a microprocessor and a hoard of memory on board.
> Think about it...or at least go review the recent discussion of
> the DPT controller for some excellent commentary, pro and con, on
> the issues involved in caching disk controllers...

I don't need to.

I talked to one of the engineers from DPT at Comdex. He explained
to me the difficulty of caching a SCSI host controller: since
the controller handles up to seven devices on the SCSI bus,
putting the cache system on the controller makes for little
improvement, if any. The best solution, and one which has
been implemented by at least one drive manufacturer, is to
put the caching components on the SCSI drive. 

When a SCSI request is sent to the drive by the controller, the
controller is free to service one of the other six devices on
the bus and the drive can check its own cache for the presence
of the data requested and respond accordingly.

The previous discussion of caching controllers pertained primarily
to RLL and ESDI formats.

-- 
Bote
Old & Improved path!: uunet!comsea!csense!bote
New & Improved path!: {zardoz|uunet!tgate|cos!}ka3ovk!media!cyclops!csense!bote

misha@aeras.UUCP (Michael Umansky) (09/28/89)

In article <22542@cup.portal.com> cliffhanger@cup.portal.com (Cliff C Heyer) writes:
>Hopefully the Mylex will give 386 land the
>*first* board that gives 900KB/sec disk I/O
>like the Amiga (at Amiga prices hopefully)
In my 25Mhz Micronics with 8Mhz AT bus and Adaptec AHA-1542A SCSI controller
and CDC WREN III SCSI drive I get about 900 KB/sec.  The Mylex controller
should deliver at least a magnitude more to be cost effective.
misha
-- 
NAME:	Michael Umansky (sun!aeras!foxy!misha)
WORK:	Arix Corp.;  821 Fox Lane;  San Jose, CA  95131
HOME:	4331 Lincoln Way; San Francisco, CA  94122
PHONE:	(408) 922-1751 (work); (415) 564-3921 (home)

palowoda@fiver.UUCP (Bob Palowoda) (09/30/89)

From article <372@csense.UUCP>, by bote@csense.UUCP (John Boteler):
> From article <19245@gatech.edu>, by ken@gatech.edu (Ken Seefried III):
>> In article <369@csense.UUCP> bote@csense.UUCP (John Boteler) writes:
>>>Curious as to how a SCSI host controller can easily perform 
>>>cache operations?
>> the DPT controller for some excellent commentary, pro and con, on
>> the issues involved in caching disk controllers...
> 
> I don't need to.
> 
> I talked to one of the engineers from DPT at Comdex. He explained
> to me the difficulty of caching a SCSI host controller: since
> the controller handles up to seven devices on the SCSI bus,
> putting the cache system on the controller makes for little
> improvement, if any.

 Did DPT do a display a real comparison of throughput at Comdex?
 Does DPT manufacture a SCSI cacheing controller?

 The only reason I mention this is if I where at Comdex showing 
 off my ESDI controller. I surely would build your confidence 
 in the product. But I don't think it a fare comparison between
 these two types of controllers.

 A better comparison of the DPT would be to that of Adaptec's
 new ESDI cacheing controller.

> The best solution, and one which has
> been implemented by at least one drive manufacturer, is to
> put the caching components on the SCSI drive. 

  I thought SCSI drives have buffers?

> 
> When a SCSI request is sent to the drive by the controller, the
> controller is free to service one of the other six devices on
> the bus and the drive can check its own cache for the presence
> of the data requested and respond accordingly.

  Won't the Mylex controller service the request from the cache?

  Another thing about the Mylex controller, it is 32 bit. 
  DPT is 16 bit I believe. 

  Something else I don't quite understand is how they sync it with
  the cpu cycle.  

---Bob

-- 
Bob Palowoda  packbell!indetech!palowoda    *Home of Fiver BBS*  login: bbs
Home {sun|dasiy}!ys2!fiver!palowoda         (415)-623-8809 1200/2400
Work {sun|pyramid|decwrl}!megatest!palowoda (415)-623-8806 2400/9600/19200 TB
Voice: (415)-623-7495                        Public access UNIX XBBS   

cliffhanger@cup.portal.com (Cliff C Heyer) (10/08/89)

bill davidsen writes...
>A) typical ESDI drives (looking a a recent spec sheet) turn 3600 rpm,
>have 32 sectors of 512 bytes. 32*512*60/1024 = 960kb max from ESDI, no
>allowance for latency, step time, etc.

Exactly. So I'm p***ed off that this is quoted in (MSDOS)sales literature, but
when you actually try a simple 10MB contig. disk to disk copy on an EMPTY 
300MB disk under DOS you end up getting 200-300KB/sec on "most"
machines.

>B) I measured the ballpark i/o of several machines by doing a dd of the
>raw device to /dev/null. Crude, but it shows about 600kb for my ESDI
>system
dido.

> From this set of reasonably useful numbers I would suggest that you
>could look for a less demanding solution in any of the following areas:
>is your application really transfer rate bound? 

When I'm copying a 120MB file and I have to wait 10 minutes instead of
2 minutes, I'm I/O bound for 8 minutes. And when you consider that I
was quoted 1.2MB/sec for ESDI and am only getting 200KB/sec I think
reason to be angry! (why should it be up to me to tune the machine
to get the "quoted" speed?)

>Do the figures you
>quoted for other machines reflect actual performance, benchmark
>performance, or hardware limits? 

Evidently it's not actual performance OR benchmark performance,
must be hardware "chip specification" (MSDOS)

>Are you asking for something which isn't being marketed?

Your figures show it IS sold because you have it...

>The only thing you implied with which I disagree is the implication
>that people will grow out of 386 type systems in the i/o end. 

People start with a 50MB database & 1 user, and in two years 
it becomes 500MB and 10 users, in three years 900MB, etc. 
If this is not growing out of a 386 system, what is?

>Cliff Heyer said previously...
>>The VAX ranges from 600KB/sec to 1.3MB/sec tops sustained I/O *per job*. Many
>>PCs I tested got in the range of 200KB/sec (ST-506 ...). Then I found to my
>>surprise that SCSI & ESDI disks on many machines was still 200-300KB/sec!  
>>This was confirmed by checking BYTE benchmarks that list 1MB throughput times.
>>SUN was the only machine in BYTE that showed a respectable 800KB/sec time.
Karl Denninger responded...
>Huh?  1.3MB/second per job, eh?
>With what interface?  And to how many jobs?  We install, service, upgrade
>and work on VAX systems all the time, and I've >never< seen one which can
>sustain 1.3MB/sec (bytes now, not bits) across more than one or two jobs!
>The typical 3xxx (or MVII) series machine with 
>ESDI or SCSI interfaces and a couple of drives?  No way!

I'm talking about one job only, such as a VAXstation 3xxx (KS650 CPU) with
a Dilog DQ246 (SMD-E)& CDC EMD 9720-750. Also get the same results
with the new SCSI Q-bus adaptor CQD-220/M & Imprimis Sabre. (all >80blk
contig. transfers single user) OK, we don't use this all the time, but if I can
spend the same amount of money for hardware that does this, why not?

Cliff wrote...
>>It has become evident to me that in spite of 15MHz ESDI chips, sync SCSI, etc.
>>certain machines still go as slow as their ST-506 counterparts. Also, unlike
>>the old mainframe days when computer were rated in actual disk I/O in KB/sec, 
>>todays micros are sold with NO ratings, except the "chip throughput" ratings
>>which are meaningless because they do not show what the actual throughput
>>is.
Karl responded...
>I don't know where you're coming from, or where you get your hardware, but
>there is no way that I can agree with this.
Cliff responds...
Have you ever benchmarked (MSDOS)386 CPUs and compared ESDI/SCSI with ST-506?
If you had, there is no way you could say this. My guess is that less than 10%
of the DOS hardware out there will give you 900KB/sec, and the rest (mail order
houses, etc.) runs the same speed as ST-506 even though the sales literature
says otherwise. I intend to study this (my degree is in hardware engineering
even though I ended up in software) to find out what they are doing.

>Every one of our RLL-based machines can do in the area of 700KB/sec raw I/O.
>Every one of our ESDI-based machines can do in the area of 900KB/sec.  What
>is lost in the OS is another story, but that's something we don't have
>control over!
>No, you can't do this with cheap and slow hardware.  You certainly CAN with
>fast and a little more expensive components.

Well I was talking about MSDOS, I know UNIX has much more overhead. I wouldn't
expect to get 900KB/sec with UNIX. Perhaps the PC-compatible makers are using
their own drivers(or something else) that ends up being slower than other machines,
or some other lower cost stuff.

>Why not buy from someone who does bother to check these things, and can tell
>you exactly what you're getting?

Good point. I guess my beef is the false advertising. Good article about this recently
in Sept. 89 MIPS P. 99 col. 1, where they compare to the 60's hifi days. The "real"
specs are all hidden and nobody knows what they are getting until after they buy
and open the cover, look at the chip numbers (if they haven't been defaced) and look
them up. 

kEITHe writes...
>Anyway, using CORE27 to test an Everex STEP/25, a WD 7000ASC and a
>CDC WREN IV (? 300 Mbytes) I got 1.04 Mbytes/sec.  That was some
>time ago, however, and I've abandon that configuration.
>
>Using CORE27 on the Intel 301, Adaptec 1542A and CDC combination I
>get 1.4+ Megabytes per second reported.  This is with no caching 
program loaded.

Cliff Responds...
OK here's someone who knows what I am talking about! BUT the INTEL 301
is big bucks right? I seem to remember that intel boards are much more
expensive (=faster DRAM etc). (comments please!)


Well back to Mylex DC376. 13MB/sec.........I have a feeling that this is with
the 4MB cache loaded and streaming out contig... How about the "raw" rate
from disk assuming what you want is not in the cache? Also, I assume this
controller is "smart" in that while it's transferring the 4th MB, it's loading
the first 3 MB with the next read it thinks you'll want so when it's at the
end of the 4th MB it will seamlessly start at the 5th.

Also, what is the AT-32? Is this their own design or is it a logical extension
of the AT bus?

davef@lakesys.lakesys.com (Dave Fenske) (10/08/89)

The "AT-32" bus, is nothing more than the bus which Intel originally used for
memory on the 301s (no folks, no Levis).  As I understand it, Intel has since
modified the specs.  As a result, the AT-32 is indeed a proprietary bus.

Your welcome.

Now, can anyone tell me what the message "123.RI is invalid" really means?

DF

zeeff@b-tech.ann-arbor.mi.us (Jon Zeeff) (10/09/89)

Forget Coretest and other low level test results regarding disk i/o 
rates.  The best transfer rate benchmark is going to be a copy from 
the block device to /dev/null with the version of Unix you are going 
to use.  

Make sure you have an unfragmented disk if you expect actual use to come
close to this figure.

-- 
Branch Technology            |  zeeff@b-tech.ann-arbor.mi.us
                             |  Ann Arbor, MI

akcs.vrona@vpnet.UUCP (David Vrona) (10/15/89)

Can you tell me how much the Mylex SCSI is?  Thanks much.