DPT Controller Summary (LONG)

Barton A. Fisk barton at holston.UUCP
Fri Oct 20 00:07:12 AEST 1989


As promised, appended below is a summary of the responses
to my inquiry regarding the DPT caching controller. The
overall response was extremely in favor of the DPT. enjoy

------Begin summary--------

On Oct 4,  9:09am, attctc!texbell!swbatl!uunet!csinc!rpeglar wrote:
} Subject: dpt controllers
} 
} Barton,
} 
} We're using them here.  Short description;
} *  intel 302 and 303 platforms (25 and 33 mhz '386)
} *  dpt PM3011/70 (esdi) controller(s) - 512k and 2mb 
} *  imprimis 383 (unformatted) esdi disk(s)
} *  SCO Xenix 2.3.2 GT and/or SCO Unix 3.2.1 GT
} 
} So far, no problems.  The machine thinks (bios) the drive is a type with
} <= 1,023 cyls.  The system (Xenix) thinks the controller is a WD (I forget
} the type offhand, more later), in order to handle the drive correctly -
} the dpt controller actually maps the OS requests to the drive's geometry.
---------------------

I installed an ESDI DPT controller and a Maxtor 650MB drive on a 386 Xenix
system and didn't have any problems. The installation was easy (the format
took a hell of a long time, not due to the controller though :-) and the
system went like the wind afterwards. The controller will vary the way the
drive looks to the system so that systems with head/cylinder/sector etc
limits that are different to the drive can still have the drive installed.

A couple of things to keep in mind :-

	- no other disk controller can exist in the system with the DPT
	  (ie throw out your old HD and HD controller!)

	- due to the above you will probably need to purchase the related
	  floppy controller (if your old HD controller also controls your FD)
	  card (under $100) which plugs onto the DPT controller

	- the DPT is available for SCSI or ESDI

	- the ESDI version will drive up to four drives

	- there is also an optional on board drive type ROM for using large
	  drives with dumb operating systems (not needed for recent Xenix's
	  but probably for DOS)

	- the cache RAM is 512KB standard, but can be expanded to several MB.


Hope this helps, dmh


Darren Hosking				ACSnet: dmh at goanna.oz	 
Department Of Computer Science		ARPA:   dmh%goanna.oz.au at uunet.uu.net
------------------
From: frankb at usource.SARASOTA.FL.US (Frank Bicknell)
Status: OR

We have one client who has the 506 variety with 4.5M of cache.
They recently had a "power problem" which managed to take out
the video card and, after much head scratching, we presume the
DPT controller, too.  (I say presume since the controller is now
at DPT being examined.)  The guys at DPT were fantastic: they are
very willing to help, even though the problem did not appear to
be with the controller at first.  Finally, after replacing
everything but the controller (well, almost :) ), he said, "you
better send it to me."   I'm glad he said that: I was about
ready to suggest the same thing.

Anyway, while it was working, it really improved their
performance.  It's a bit expensive, but a very nice product.

Another client we are fitting with the ESDI + mirror module
variety (no extra cache).  That unit isn't in the field yet, of
course, but it works great so far.  We in fact had one of the
drives which was DOA, and the mirror module reported it "dead"
just as it was supposed to.
-- 
Frank Bicknell
UniSource; 1405 Main St, Ste 709; Sarasota, FL 34236
attctc!usource!frankb || frankb at usource.SARASOTA.FL.US

---------------------------
From: garyb at gallium (Gary Blumenstein)
Status: OR


I'm using one of these beasts with 4MB of cache memory and two Seagate
4096's (standard ST412 interface).  I wouldn't trade this package
for ESDI as far a useable speed under load.  The only reason why I'd go to
ESDI is if I needed more capacity.   This setup is great.  All you have
to do to be convinced is log in here while rnews and compress is munching
away at 60 or so 50K news batches!

I have some crude benchmarks which I ran here and I'll send them to you
when the sun comes up here and I'm not as "zoned" as I am now at this
ungodly hour!  ;-Q
-- 
From: attctc!texbell!swbatl!uunet!csense!bote (John Boteler)
Status: OR

Subject: Re: DPT's caching controller

A client is using one.

They are really only useful if you intend to use the host computer
as just that--a host for many users; otherwise it is a waste of
money for a single user machine.

Having said that, the DPT can make a substantial difference
in performance. I refuse to quote numbers because a) I haven't
any, b) the ones posted to the net conflict, c) I haven't
seen any benchmarks which can get a handle on how much
improvement is obtained, even ballpark figures.

The nicest parts of the DPT design is never noticed by the
user. When installing the board, its supplied installation 
program formats the drive in an optimum manner for the
operating system of choice. For example, Xenix gets its
sectors skewed by a calculated percentage in order to
maximize the efficiency of data I/O. 

The cache itself is not battery backed, but it flushes to
disk after about 0.25 second of inactivity. The advantage to
a hardware cache is that even if the machine crashes or
locks up, after that 1/4 second the cache will write itself
to disk and an orderly shutdown may proceed. The disadvantage
someone pointed out is that UNIX makes certain assumptions
about the order in which data is written to disk, and
the DPT system breaks those assumptions. I don't know
if this is critical or not. Since the DPT uses an 
elevator seeking algorithm to avoid a lot of head thrashing,
this sounds true at first blush. I do know that my client had
some mysterious data corruption problems on their system
at first, but when I pressed their genius on the matter
I discovered he was shutting the system off cold because
of another bug in the kernel, not because the DPT was
simply trashing data.

It is sometimes difficult to separate fact from fantasy.
----------------------------

From: attctc!texbell!wuarchive!ncar.UCAR.EDU!jerry%xroads%mcdphx%asuvax (Jerry M. Denman)
Status: OR


I have nothing but good things to say about the DPT Controller.  It
has been a real life saver on systems with 15 or more users.  I currently
have them installed with 6 users and not one has given me any problems.

The performance increases are fabulous when combined with the Anvil Designs
serail I/O controller.  I have a large system ( 40+ logins, 660 Mb of disk,
16 Mb of memory  ) that the response time is equal with 1 or 40 users.

If you can justify the expense then buy it.

---
Jerry M. Denman
Director of Technical Services
Peak Systems, Inc.
Phoenix, Arizona
(602) 397-UNIX
-- 
Barton A. Fisk          | UUCP: {attctc,texbell,vector}!warble!holston!barton
PO Box 1781             | (PSEUDO) DOMAIN: barton at holston.UUCP     
Lake Charles, La. 70602 | ----------------------------------------
318-439-5984            | "Let him who is without sin cast the first stone"-JC



More information about the Comp.unix.xenix mailing list