[comp.sys.dec] Can a DECstation replace a 11/750 ?

aks@somewhere.ucsb.edu (Alan Stebbens) (05/04/90)

gordon@prls.UUCP (Gordon Vickers) writes:

>       I'm considering replacing our VAX 11/750 (Ultrix) with something a bit
>   cheaper to maintain.  Our DEC sales rep. is suggesting a DECstation 3100
>   or the newly announced DECstation 5000.

We replaced our VAX/780 running 4.3BSD with a DECstation 3100 running
Ultrix 2.2.  Since our original computer was a CPU resource, we do not
use the console as an X terminal, which offers more memory for other
applications.

>      Can such a little box really replace a 750 ?  I know technology has
>   come a long way, but still ....... that's an awful small box.

Yes.  Size has nothing to do with speed.  The DS3100 gets about 15 MIPS.
What does a VAX get: 1 MIP.

>      Our 750 seldom has more than four user's but I manage to bog it down
>   with an applications program that runs continously.  This program reads
>   and writes to nine serial ports that communicate over LADS lines at
>   2400 baud to remote equipment. The incoming data occurs in bust of about
>   one kbyte (at random times) , typicaly no more than two ports receiving
>   at a time. We collect about one megabyte of data per day but this will
>   increase to about three megabytes per day within the next two years.

The DECStation is limited to two serial ports, but we bought a terminal
server which then connects to the DS3100 via Ethernet.  All of our users
connect to the terminal server via hard-wired lines or modems, and then 
login to the DS3100.  Also, the terminal server allows connections to our
other systems, and we do not have to provide more wiring.

>      We use two disk drives, an RA80 and an RA81 -Yuck!  We really need
>   something faster and at least slightly more disk space. Something
>   equivilant (in capacity) to two RA81's would work (I don't like the
>   partitioning options of the RA81 either).

We added SCSI CDC Wren VI's to our DS3100 and ran some simple disk
timing programs which showed that the I/O timings were the fourth
fastest in our entire machine room, including SMD drives on SMD
controllers on Sun servers.  Because the DS3100 is fast, it's SCSI
driver is fast, and the new SCSI drives from CDC or Hitachi are very
fast (15-18 ms).  In fact, the new, external drives were faster than
the DEC-supplied internal disk drives.

Also, the cost of the Hitachi 700MB drive was about $2800.  So, for about
$5600 you get 1.5MB of disk space.  Compare this to the cost and size of an
RA80 or 81!

>     The DECstation literature and catalogues I have say nothing about
>   how one adds serial ports.  Are serial devices assumed to be connected
>   via a terminal server or can serial ports be added through an inteligant
>   controller board ?

You can't expand serial ports (I believe); however, why do this when there
are many manufacturers of decent terminal servers which do the job even
better.

>     Can either of these systems realistically support eight simutanious
>   users ?  These are such small boxes, it reminds me of a mulituser
>   PC advertisement I once saw that claimed to support 32 users.  I'm
>   spoilled, I want a computer that waits on me, not visa versa.

Again.  It appears that you are overly concerned with size, and not
performance.  The DS3100 really runs at 15 MIPS, which is *fifteen*
times faster than a VAX/780.  If a VAX/780 could support 8 users, why
can't a machine which runs faster?  Again, size has absolutely nothing
to do with performance.

>     Any followup would be appreciated. I'm not neccessarily looking for
>   a "Yes/No it will do the job" type reply. At this point any information
>   would be welcomed.

The things which you should be concerned with are not the performance
issue, which is quite clear, but the compatibility issue, which is not
at all clear.  There were many applications which we had to port from
the VAX/BSD environment to that of the DS3100/Ultrix system.  This
required time, and, in fact, some ports didn't even work.  For example,
ditroff was never able to port sucessfully (the standard ditroff from
Berkeley).  We had to buy a commercial package to get roff
functionality.

One area which we saved really big was in maintenance.  The cost of
maintaining a VAX/780 was tremendous, and we could buy about two
DS3100's a year just on the savings alone.  This means, really, that we
don't have to get maintenance for the DS3100 -- if it breaks, we buy a
new one, and we're still ahead of the maintenance costs.

Alan Stebbens        <aks@hub.ucsb.edu>             (805) 961-3221
     Center for Computational Sciences and Engineering (CCSE)
          University of California, Santa Barbara (UCSB)
           3111 Engineering I, Santa Barbara, CA 93106

Alan Stebbens <aks@hub.ucsb.edu>

eric@spock.UUCP (Eric Volpe) (05/04/90)

In article <5070@hub.ucsb.edu> aks@somewhere.ucsb.edu (Alan Stebbens) writes:
>
>Again.  It appears that you are overly concerned with size, and not
>performance.  The DS3100 really runs at 15 MIPS, which is *fifteen*
>times faster than a VAX/780.  If a VAX/780 could support 8 users, why
>can't a machine which runs faster?  Again, size has absolutely nothing
>to do with performance.

It is *not* *fifteen* times faster than a vax 780! the DS3100 is a RISC
machine, meaning Reduced Instruction Set - Which means that each instruction
on a RISC processor is very small and the equivalent of several instructions
of a non-RISC processor. So, while it may be true that the DS3100 can
execute fifteen times as many of its instructions in one second as a VAX780
can of vax instructions, you're comparing apples and oranges. Each Vax
instruction accomplishes perhaps the equivalent of eight DS3100 instructions.

I'm not saying that the DS3100 isn't faster than a vax780 anyway, I'm just
saying that you can't blindly quote MIPS ratings and expect it to truly
reflect the relative performance of two totally different architectures. 

Also, Instructions per Second is not a complete description of the power
of any computer - there are many other considerations; for example, Bus
I/O speed, transfer rate, caching, etc. I'm not sure what kind of bus the
DS3100 has (is it Qbus? I haven't gotten mine yet) but the Vax780 has a very
fast bus indeed, and as far as handling many terminals, a fast bus is far
more important than a fast processor - my PDP-11/34 can handle 8 users
better than a fast DOS machine with an 8 terminal board, though the AT 
processor is indisputably much faster. So you have to look at many factors
to be able to judge the relative power of two different processors. 

						Eric Volpe


-------------------------------------------------------------------------------
Eric Volpe					  ...uunet!hsi!yale!spock!eric
Choate Rosemary Hall '90			[.sig still under construction]
-------------------------------------------------------------------------------

grr@cbmvax.commodore.com (George Robbins) (05/04/90)

In article <1990May4.025422.8656@spock.UUCP> eric@spock.UUCP (Eric Volpe) writes:
> In article <5070@hub.ucsb.edu> aks@somewhere.ucsb.edu (Alan Stebbens) writes:
> >
> >Again.  It appears that you are overly concerned with size, and not
> >performance.  The DS3100 really runs at 15 MIPS, which is *fifteen*
> >times faster than a VAX/780.  If a VAX/780 could support 8 users, why
> >can't a machine which runs faster?  Again, size has absolutely nothing
> >to do with performance.
> 
> It is *not* *fifteen* times faster than a vax 780! the DS3100 is a RISC
> machine, meaning Reduced Instruction Set - Which means that each instruction
> on a RISC processor is very small and the equivalent of several instructions
> of a non-RISC processor. So, while it may be true that the DS3100 can
> execute fifteen times as many of its instructions in one second as a VAX780
> can of vax instructions, you're comparing apples and oranges.

Please, let's avoid the RWARS...

As soon as I got my 5810 up, I ran some dumb benchmarks, like some monster
diffs that were painfully slow on our 785.  I believe the performance ratio
was 9:1 and if you apply the standard 785:780 factor you get about 14 VAX
mips.  The 3100 is maybe two thirds of this, so you're in the 10 VAX mips
range.  Your milage varies depending on your driving habits.

Obviously this isn't 15 times 780 performance, on the other hand, a bunch
of RISC MIPS vs. VAX MIPS assertions don't exactly knock it back to 780
levels either.  I/O performance is obviously another issue, but for an
8-user system, you've probably got nothing to complain about, as long as
you invest in enough memory to avoid wasted cpu/disk resources for paging. 

-- 
George Robbins - now working for,     uucp:   {uunet|pyramid|rutgers}!cbmvax!grr
but no way officially representing:   domain: grr@cbmvax.commodore.com
Commodore, Engineering Department     phone:  215-431-9349 (only by moonlite)

beaulieu@gca.UUCP (Larry Beaulieu) (05/04/90)

	I'll preface by commenting that we have both an 11/780 (Ultrix 3.1, 3 x RA81,
	2 x Unibus, TU77's, etc. etc.) and have had a DS3100 (Ultrix 3.0, 322MB drive, TK50 - ugh! ) 
	for a year.
 
> It is *not* *fifteen* times faster than a vax 780! the DS3100 is a RISC

	True.  On some of my applications it is 30 times faster :-)

	MIPS ratings are usually based upon timing tests performed on 1 or 2
	specific test suites such as the infamous Dhrystone & Whetstone tests.
	For the DS3100, the test they selected obviously runs 14 times faster
	than on an 11/780, ergo the sales rep spouts ITS A 14 MIP MACHINE.

	My perception of reality is that it averages to about a 9 MIP machine,
	a tad faster than the SparcStation 1 (but what about the SS1+? :-),
	as normalized to our 11/780.

> execute fifteen times as many of its instructions in one second as a VAX780
> can of vax instructions, you're comparing apples and oranges. Each Vax
> instruction accomplishes perhaps the equivalent of eight DS3100 instructions.
	
	However, a RISC box would execute its 8 instructions a lot
	faster than the 780 would execute its 1 instruction.
 
> I'm not saying that the DS3100 isn't faster than a vax780 anyway, I'm just
> saying that you can't blindly quote MIPS ratings and expect it to truly
> reflect the relative performance of two totally different architectures. 

	Yup.  SPEC or DR ratings are even better (although still not great), 
	but the best thing anyone could do is to try before you buy, if 
	at all possible, with your own applications.
 
> Also, Instructions per Second is not a complete description of the power
> of any computer - there are many other considerations; for example, Bus
> I/O speed, transfer rate, caching, etc. I'm not sure what kind of bus the
> DS3100 has (is it Qbus? I haven't gotten mine yet) but the Vax780 has a very
> fast bus indeed, and as far as handling many terminals, a fast bus is far

	SCSI.  The Decsystem 5400 uses the Q-bus as the I/O bus.  

	The throughput ratings of the 2 buses are similar.
	
	SCSI - 4 MB/sec asynchronous, 5 MB/sec synchronous (theoretical)
	Q-bus  5 MB/sec (theoretical), ~3.3 MB/sec actual maximum throughput
	Unibus - ~2 MB/sec (theoretical), ~1.2 MB/sec actual maximum throughput

	The 11/780 uses the Unibus for the Ethernet interface and for the RA series
	drive controllers.  

	I'll defer on the Massbus, since only the older hard drives (RK, RM series)
	use it directly, and their performance leaves much to be desired, even
	against the RA81.

> more important than a fast processor - my PDP-11/34 can handle 8 users
> better than a fast DOS machine with an 8 terminal board, though the AT 

	If I were single streaming batch jobs, I'd take the AT any day.

	You are absolutely, positively, correct if you are talking about
	an interactive environment.  We get better interactive thoughput
	from our 11/84 (.72 MIPS) running RSX than we do from our 11/780.

	However, at least part of this discrepancy is probably due to
	the relative sizes and efficiency of the RSX executive vs. the
	Ultrix kernel.

	The odds are pretty good that a fast processor will keep a slow bus
	working at close to its maximum bandwidth.

	To conclude:

	In my experience the 780 is a slug compared to the DS3100, and is
	much more expensive to keep around.

	I remember an article published in the DECUS newsletter last year
	about someone porting their application to a DS3100 in order to
	do a performance evaluation; as I recall, they ported some of their
	simulations written in Fortran.

	In comparing the DS3100 to their VAX 8810:

		The DS3100 smoked the 8810 big time for compute-intensive
		applications.

		The 8810 offered 2x the speed of the DS3100 for I/O intensive
		applications.

	However, they were comparing a $500,000+ machine vs. a $30,000
	one.
	
 
-- 
Larry Beaulieu 			When up to your a** in alligators, it's
SMTS/Software Engineer		difficult to remember that your original 
GCA Corporation, 		objective was to drain the swamp...
Andover, MA	...uunet!gca!beaulieu	The usual disclaimers apply.

rogerk@mips.COM (Roger B.A. Klorese) (05/05/90)

In article <1990May4.025422.8656@spock.UUCP> eric@spock.UUCP (Eric Volpe) writes:
>In article <5070@hub.ucsb.edu> aks@somewhere.ucsb.edu (Alan Stebbens) writes:
>>
>>The DS3100 really runs at 15 MIPS, which is *fifteen* times faster than a 
>>VAX/780.
>It is *not* *fifteen* times faster than a vax 780! the DS3100 is a RISC
>machine, meaning Reduced Instruction Set - Which means that each instruction
>on a RISC processor is very small and the equivalent of several instructions
>of a non-RISC processor. So, while it may be true that the DS3100 can
>execute fifteen times as many of its instructions in one second as a VAX780
>can of vax instructions, you're comparing apples and oranges. Each Vax
>instruction accomplishes perhaps the equivalent of eight DS3100 instructions.

If "MIPS" [as opposed to "Mips" ;-)] had a single useful meaning, we could
determine that one of you is right, and the other, wrong.

But I'd have to say that Eric is "wronger" than Alan.

First, the MIPS spoken of here is not the execution of actual, machine
level instructions.  It is the "multiple of <some benchmark> compared
to a VAX 11/780" measurement.

If the benchmark is Dhrystone, the ratio of 15:1 is approximately correct;
this is why DEC announced the machine as 14-15 "Dhrystone MIPS," or
"integer MIPS," or some such.

If the benchmark is the SPEC suite, the number is lower, but I remember it
as 10x, +/- 2x.

If the benchmark is the Digital Review benchmark suite, or the DEC
performance suite, or the Mips performance suite, the number is, if I
remember correctly, in the 12 range.  Both DR and DEC tend to call these
VUPs, or "VAX-equivalent units of processing."

Of course, the only meaningful system benchmark is your own actual 
workload.

But, on the other hand, it is a crock that there is an 8x instruction ratio
in most code.  Our static comparisons yield a typical ratio of about 1.4x;
dynamic comparisons vary with the code, of course, but it is never (well,
hardly ever) observed as anywhere near the 8x range.

Actual MIPS, that is, how many instructions get executed in a second, is
a measurement that can be viewed in the context of architectural efficiency,
or in the measurements for certain specific applications, like limited
embedded tasks.  But it is so rarely useful that most people do not use
it as the only number in their spec-sheets unless they have something to hide.

These VAX/RISC comparisons are *not* true MIPS, but rather, integer program
ratios, which are much more (if not completely) relevant.
-- 
ROGER B.A. KLORESE      MIPS Computer Systems, Inc.      phone: +1 408 720-2939
MS 4-02    950 DeGuigne Dr.   Sunnyvale, CA  94086   voicemail: +1 408 524-7421 
rogerk@mips.COM         {ames,decwrl,pyramid}!mips!rogerk         "I'm the NLA"
"Two guys, one cart, fresh pasta... *you* figure it out." -- Suzanne Sugarbaker

mash@mips.COM (John Mashey) (05/05/90)

From the currently best available SPEC benchmarks:

The DS 3100 is: 11.8X faster than VAX-11/780 on 4 SPEC integer (geometric mean)
		10.9X faster ... on 6 SPEC FP
		11.3X faster on the entire 10-benchmark mix (SPECmark)

These programs have little I/O or kernel time, os these are CPU benchmarks.

The least ratio is 9.5 (spice) and the greatest is 13.2 (nasa7).

The Sun SS1+ has a published SPECmark of 10.0; I don't havethe individual
numbers, but I'd expect it to be about equal on integer, and somewhat
slower on FP. (Based on clock ratios from SS1).

The benchmarks are big enough to thrash the caches around at least some;
which gives the 780 a fairer break.  [Small benchmarks fit in most RISC
caches, and so they go faster than a VAX relative to how they do on more
realistic ones.]
-- 
-john mashey	DISCLAIMER: <generic disclaimer, I speak for me only, etc>
UUCP: 	 mash@mips.com OR {ames,decwrl,prls,pyramid}!mips!mash 
DDD:  	408-524-7015 or 408-720-1700
USPS: 	MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086

lindh@uhasun.hartford.edu (Andrew Lindh) (05/05/90)

In article <38515@mips.mips.COM>, rogerk@mips.COM (Roger B.A. Klorese) writes:
> In article <1990May4.025422.8656@spock.UUCP> eric@spock.UUCP (Eric Volpe) writes:
> >In article <5070@hub.ucsb.edu> aks@somewhere.ucsb.edu (Alan Stebbens) writes:
> >>
> >>The DS3100 really runs at 15 MIPS, which is *fifteen* times faster than a 
> >>VAX/780.
> >It is *not* *fifteen* times faster than a vax 780! the DS3100 is a RISC
> >machine, meaning Reduced Instruction Set - Which means that each instruction
> >on a RISC processor is very small and the equivalent of several instructions
> >of a non-RISC processor. So, while it may be true that the DS3100 can
> >execute fifteen times as many of its instructions in one second as a VAX780
> >can of vax instructions, you're comparing apples and oranges. Each Vax
> >instruction accomplishes perhaps the equivalent of eight DS3100 instructions.
> 
> ........ 
> Of course, the only meaningful system benchmark is your own actual 
> workload.
> 

Yes it can replace a 750....in some cases...
The CPU may be able to get more done, but there is more to a VAX than just
the raw CPU speed.  Yes it may blow away a 750 if you are doing a lot of
number crunching with a small number of users. But I don't think that
the disk are other equipment is as fast. But I do know our new 6320 is 
faster than our old 780. At 40 or 50 users it (6320) is still faster
than 20 people on a 780...and the 6320 has 6Mips vs. 3Mips (780)
(here we are back to the MIPS stuff....but they are the same CPU)

As an example (Non DEC stuff)....I was looking at the new Sun SparcStations
and the new IBM RT/6000 stuff. The IBM had a faster CPU but it seemed
to be a MUCH slower machine in terms of the user interaction and graphics IO.
--
Andrew Lindh, a student at the University of Hartford -- Computer Science
BITNET:    LINDH@HARTFORD.bitnet       INTERNET: lindh@uhasun.uofh.edu
UUCP/Usenet: lindh@evecs.uucp	---- When will I grduate???
NOTE: All views here are MINE!!! Not the schools or thoes of anyone else!