lien@osu-eddie.UUCP (02/10/87)
From: Yao-Nan Lien <lien>
Two weeks ago I sent out a message to ask for the experience of
Encore and Sequent systems.
Here is a list of all of the responses I received.
Many thanks to everyone who responded and for all of the
useful information received.
We having make decision yet. The comparison is rather hard
since both system look good. We will keep to collect more
information to help us to make final decision.
Please send me a mail if you have more information.
The summary is divide into four sections.
1. The messages that compare both systems.
2. The messages about Sequent system
3. The messages about Encore system
4. Others
------------------------------------------------------------
Yao-Nan Lien
Department of Computer and Information Science
Ohio State University
2036, Neil Ave. Mall
Columbus, Ohio 43210-1277 (614) 292-5236
CSNet : lien@ohio-state.CSNET
Arpa : lien@ohio-state.arpa
UUCP : {cbosgd, ihnp4}!osu-eddie!lien
-------------
1. COMPARISON
-------------
---------- ihnp4!lll-lcc!esl.ESL.COM!ted (Thomas DuBois)
We are currently involved in making a similar decision between Sequent, Encore
and Alliant. The Alliant has quite an interesting architecture but without
knowing your needs I don't know if its appropriate.
We are looking for a machine to satisfy several groups here at esl. There is
a need for a lot of number crunching ability, Ada support, and a desire
to model distributed systems and/or parallel architectures on a parallel
machine. The included report was compiled based on these criteria.
We have decided to go with the Alliant if funds are available, the Sequent if
not. Encore has a much less credible record, though in an academic atmosphere,
lack of software may not be a key issue.
Hope this is helpful. Go Buckeyes!
Tom DuBois
vecpyr!esl!ted
Parallel Computer Comparision Matrix
The following data is the result of comparing four computers,
Sequent's Balance System, Encore's Multimax, Alliant's FX series,
and the new Sun4. The first three machines are parallel computers
with similar architectures. The Sun4 was included as a conventional
alternative.
In order to immediately satisfy the diverse requirements of several groups in
AT, the Sequent machine is the computer of choice. They are the only vendor
currently able to fulfill all of our key requirements, namely supporting
Ada, number-crunching applications , and parallel system simulation at a
reasonable price. Their system is priced competitively, they have an
excellent reputation, and their machine can expand as our needs change.
Performance
Sequent - 2.8 - 21 MIPS, Next generation 50Mips
Encore - 1.5 - 15 MIPS, Next generation 40Mips, 40Mwhetsone
Alliant - 94.4 MFLOPS, 35 MIPS with 8 CEs
Sun4 - 8-10 MIPS
Configurability
Sequent - 26 slots
4 to 30 microprocessors(2 CPUs/board)
Up to 6 8Mbyte Memory boards w/ controllers
Up to 4 Ethernet interfaces
Disk Controller
Up to 44 Multibus slots
Encore - 20 slots
System Control Card (1 slot)
2 to 20 microprocessors(2 CPUs/board) (11 slots split
w/Ethernet cards)
4-128 Mbytes shared memory cards (up to 8 slots)
Ethernet/Mass storage cards (11 slots split w/ cpu cards)
Alliant - 8 Complex computational processors(gate arrays w/ FP
processors)
12 interactive processors (68012
4 32Kbyte IP caches
128 Kbyte interleaved CE cache
8 8Mbyte shared memory modules
Crossbar interconnect
Up to 12 independent multibus I/O systems
CPU
Sequent - 32032, Second generation CPU?
Encore - 32032, 32332
Alliant - CEs are proprietary vector processors, IPs are 68012s
Sun4 - RISC based proprietary CPU
Floating Point - All IEEE format
Sequent - National floating point co-processor currently, weitech on
second generation
Encore - National floating point co-processor currently, weitech
coming, demonstrating systems now
Alliant - Pipelined vector/floating point CEs
Interfacing
Sequent - Multibus interface, Ethernet interface, Foundation board,
SCSI interface
Encore - Ethernet, SCSI,or user-designed nanobus adaptor
VMEbus adaptor expected Nov. 1
Alliant - TCP/IP, multibus, DR11-W
Sun4 - VMEbus
Bus Technology
Sequent - SLIC bus for interrupts and synchronization, main bus is
26.7 Mbyte/sec sustained bandwidth modified multibus
Encore - 100 Mbyte/sec nanobus, deferred response, pipelined bus
interfaces.
Alliant - 4 X 8 Crossbar, 376 Mbyte/sec sustained bandwidth
Four way interleaved, 150 Mbyte/sec memory bus
Sun4 - Standard VME, not a multi-processor
Memory
Sequent - 48 Mbytes
Encore - 4-128 Mbytes, 8 way interleaving, 100 Mbyte/sec.
128 Mbytes achieved using 1 Mbyte drams, expensive
Alliant - 64 Mbyte
Sun4 - 16 Mbyte
Disk
Sequent - SMD interface, Dual Channel Disk Controller
Encore - SCSI interface, with huge buffer cache. Possible
problem with losing data
Alliant - SMD, 8kbyte fifo
Sun4 - SMD
Array Processor
Sequent - Mercury on Multibus, Mercury will provide device driver in
March
Encore - Will give us bus specs, VME adaptor being designed
We will have to write driver
Alliant - Vector machine already
Sun4 - Sky Warrior ($3000)
Unix
Sequent - 4.2, V separate or together versions, praised by
references (Bob Brown, NASA Ames) as very stable and complete compared
to Sun and Pyramid.
Encore - 4.2, V versions of parallel os available
Alliant - 4.2 parallel os available
Sun4 - 4.2
Ada
Sequent - Verdix, supported by Verdix and Sequent.
Encore - Verdix Ada
Alliant - Verdix Ada, released in < month
Sun4 - Verdix Ada
Parallel Ada
Sequent - March, 87 release in test now. Used at Univ. of Colorado,
highly recommended by them. Supported by Sequent.
Encore - Released in August
Alliant - Coming within a year
Sun4 - NA
Fortran
Sequent - Yes
Encore - Yes
Alliant - Yes, vectorizing
Sun4 - Yes
Parallel Fortran
Sequent - Yes
Encore - Yes
Alliant - Yes
Sun4 - NA
Support Tools
Sequent - Parallel debugger, 1 us timer board option
Encore - 1 us timer, software tools are weak, basic primitives
provided
Alliant - Parallel debugger, 1 us timer
Sun4 - Sytem 10ms timer
Ease of Parallelism
Sequent - Maximum of 32,000 semaphores, os handles process to
processor assignment, priorities determine execution order
Encore - Unlimited number of semaphores, os handles process to
processor assignment
Alliant - Unlimited semaphores, Can configure multiple CE's as 1 large
CE, can specify processes to processors if desired
Synchronization in HW, not SW
Sun4 - None
Reliability
Sequent - Both references said reliability was very high. One has
never had a system crash. Auto-reconfiguration
Encore - Self-test code included. Auto - reconfiguration
Alliant - Auto-reconfiguration, special diagnostic bus
Sun4 - Beta test
Initial Purchase Cost
Sequent - $421K for 24 CPUs, 24 Mbytes, 1056 MB Disk, Tape Unit,
Ada, Fortran, C compiler, Parallel Debugger
$404K, for 10 Second generation CPUs in above configuration
Encore - $317K, for 1st configuration with 32032 and no weitech
$569K, for configuration above with 32332 and weitech
Alliant - $300K for 2 CE, 2IP useable system
Discount available if involved in distributed research
Greater than $1M for loaded system
Sun4 - ~80K
Cost cpu board
Sequent - $20K (Dual microprocesor)
$35K (Second generation w/FPA)
Encore - $29K (Dual micro w/o weitech)
$39k (32332 w/ weitech)
Alliant - $49K (Complex processor ~ Vax8600)
Sun4 - NA
Maintenance Cost
Sequent - $4749/mon
Encore - $2773/mon
Alliant -
Sun4 -
Third party support program
Sequent - Yes
Encore - No
Alliant - No
Sun4 - Yes
Company Stability
Sequent - Founded 1984,
Market value of $75 million
Encore - Founded 1983, Beta 12/85. Losing $2-4 million quarter.
Market value of $15 million
Alliant - Founded 1982,
$30 million revenue last year
Company base
Sequent - Installed base of 150 systems
Encore - 38 systems installed
Alliant - Installed base of 100 systems
Sun4 - Beta-test
Comments:
Sequent - Flat response time, well supported machine. Haven't
seen any problems with bus contention. Extremely
easy to use, transitioned department from Vax
to Sequent with no major problems - Bob Brown
Encore - Very few encore systems in use, look more stable than
they did a year ago however. Supposed to beta
test multi-max but release was pushed back until
they gave up trying. - Bob Brown
Sequent - Machine has never gone down. Sequent is far ahead of Encore
in Ada area. Very easy to use. - Peter Hoffman
Encore - Best machine from hardware viewpoint, but system software
and support is lacking. - Peter Hoffman
Alliant - Outside of university price range. Model of vectorization
is a problem, different than cray. Also bus congestion
occurs for certain vector problems. - Peter Hoffman
-------------
2. SEQUENT
-------------
-------- abe <abe@J.CC.PURDUE.EDU>
We have a Sequent B21K and a B8k which are used for instructional
computing. We are very pleased with them. Porting existing 4.2
and 4.3 programs from VAX-11/780s is very easy. The hardware is
very reliable and very well designed.
Our B21K has 12 processors, 24MB of memory, 144 ports, and 3
Eagles on ZDC controllers.
There is a Sequent news group, comp.sys.sequent, I think.
Via Abell
---------- kumar@ut-sally.UUCP (Vipin Kumar)
We have had very positive experience with Sequent that we baught
more than a year ago. We have been using that as a unix machine.
We have recently acquired another Sequent primarily for
parallel processing research. My students have had very positive
experience with Sequent as a machine for parallel programming.
Vipin
---------- abe@J.CC.PURDUE.EDU
DYNIX 2.1 is a good UNIX implementation and is well documented.
If you can handle DEC software, Sequent UNIX should be no problem at all.
Vic Abell
---------- hds%pp@mcc.com (Herb Schwetman)
We purchased a B8000 in Dec. '85. It was installed in Jan. '86 and given
almost perfect service since then. Our system has 12 processors and 16
mbytes. We use it both to provide UNIX cycles to part of our research
staff and also as a tool for doing research in parallel processing.
Our only problem was with the Ethernet interface, and Sequent has worked
on this and things are fine now.
If you have questions or need further info, call or write:
Herb Schwetman (hds@mcc.com)
MCC
3500 West Balcones Center Drive
Austin, TX 78759
(512) 338-3428
---------- <krall%pp.mcc.com%pp@mcc.com (Ed Krall)>
We in Parallel Processing at MCC bought a 12 processor SEQUENT Balance
8000 about a year and a half ago. We are well satisfied with it
and are thinking about upgrading.
Problems: Sequent has had some problems with the ethernet boards. They
seem to have ironed the bugs out of the system. (Keep in mind
that we have a large and complex TCP/IP network at MCC, with all
kinds of equipment on it.)
Uses: The Sequent provides us with two distinct classes of usage: (1)
more compute power in a typical UNIX time-sharing environment, and
(2) hardware on which we can and have implemented parallel
processing languages.
Compatibility: No obvious problems with BSD 4.2.
If you need more information, I can get you in touch with people who have
more details on system internals.
-Ed Krall
Parallel Processing Program
Microelectronics and Computer Technology Corporation
3500 West Balcones Center Drive
Austin, Texas 78759
(512) 338-3406
ARPA: krall@mcc.com
UUCP: {ihnp4,seismo,ctvax}!ut-sally!im4u!milano!mcc-pp!krall
---------- Martin Jones <mcvax!cs.qmc.ac.uk!martin>
We have a Sequent Balance 21000 with 6 processors and 16Mb.
We use it as a general purpose time-sharing machine for academic
staff (about 40 users at present). We previously had a
Balance 8000 with the same number of procs but only 8Mb.
I am system administrator for this machine, and my overall
inpression is excellent. I am very impressed by the
performance and software and hardware reliability of the
machine.
As a company, Sequent Europe (who are very closely tied to
Sequent USA) are very helpful and provide very good support.
(Not that we need them very often).
I heartily recommend Sequent machines to anyone who wants a
powerful and expandable multi-user machine at a reasonable
price.
Martin Jones.
Martin Jones
UUCP: martin@qmc-cs.uucp or ...seismo!mcvax!ukc!qmc-cs!martin
ARPA: martin%cs.qmc@cs.ucl.ac.uk Post: Dept of Computer Science
JANET: martin@uk.ac.qmc.cs Queen Mary College
Easylink: 19019285 University of London
Telex: 893750 QMCUOL Mile End Road
Fax: +44 1 981 7517 London E1 4NS
Voice: +44 1 980 4811 x3933 England
---------- ihnp4!verdix!sbq (Sam Quiring)
We have (and are) worked closely with Sequent and have their Balance
8000 computer. Mechanically, electrically, and software-wise, the
Balance series is ultra reliable. The machine is fairly fast, with
each processor delivering about a Vax 750 in performance. The parallel
processors make it a great time-sharing computer as the load stays
constant even after many people have logged on.
Our business is Ada compilers. To call your compiler "Ada" you must have
it validated by the DOD. This validation consists of 2600 (or so) tests
partitioned into chapters according the the Ada Language Reference Manual.
When we validated our compiler on the Balance 8000, we had the computer
fully populated with processors (12 as I recall) and memory (16 MB).
We were able to validate in just over 4 hours by running chapters
in parallel. Our previous record for validation had been a Vax 8600
which took us about 6 hours. At that time, a Vax 750 took a little over
24 hours to validate.
They support two kinds of parallel processing. One is that each Unix
process can be run on a different processor in parallel. It is also
possible to implement parallel algorithms inside a single Unix process
and have THE SAME Unix process can be simultaneously executed on multiple
processors. This 2nd technique is very nice in some circumstances since
you can avoid the cumbersome inter-processes communication mechanisms
of Unix and use simple shared memory, all in one C program.
I can't say anything about Encore or their computers personally, since I
have not ever worked on their computer.
Sam Quiring
Verdix Western Operations
ihnp4!verdix!sbq
-------------
3. ENCORE
-------------
---------- north@tcgould.tn.cornell.edu (Michael J. North)
Yao-Nan,
I saw your note in comp.arch.
Though I don't know too much about the Sequent System, I did work
on the Multimax at The University of Oklahoma (uokmax).
Overall the Encore is a good system but it has its flaws.
On the plus side, the support we received from Encore was good.
They seem very interested in having a good reputation for response.
It handles quite a few people (we had something like 85 people logged in
at a load average of about 10-12) and response didn't seem to degrade.
The machine had as many as 97 users on it at peak times.
Getting the Annex boxes with the Encore would be advisable. They are
part of the reason the load isn't very high, all those gettys don't
need to run for the most part. Also, connecting any of your other
UNIX machines is fairly simple (if wired correctly). The printers
and modems that are there are also connected to the annexes.
I think the total cost of a terminal was about 350-500 dollars.
That included the terminal and all the wiring and labor that had
to go into it. I'm not sure about the exact price but I am rather
sure that is the range for it.
Compilers and Software:
The C compiler is the best thing on there. It produces very
reliable code. Fortran and especially !*Pascal*! have a number of
problems. The problems are diminishing as Encore puts out more
releases, but as one of the Student Applications Analysts I was
usually bombarded with questions about why things were not
working correctly. The tendency is that alot of the compilers
may be "greenware" instead of software. C-Prolog works real well.
You may wish to keep in mind that Prolog uses quite a bit of memory.
Enough to the point that when I left Oklahoma, John Hawley
(ihnp4!okstate!uokmax!hawley) our manager , was I believe evaluating
getting more memory. The problem was bad enough that when 10 people
or more ran prolog, there would be quite a bit of swapping going on
and occasionally commands typed in from the shell were being denied.
Miscellaneous oddities:
When we first got it in late June, we beat the hell out of
it to see what it could handle. Writing an 8 line C program which
makes recursive memory allocations could crash the system in about
10 seconds. This may have been fixed in the kernel by Encore
(I don't know) but it is sonmething you may want to know about because
it (crashing the system) can apparently be done quite easily.
The Encore has a very tough time getting the tape drive to
stream properly at 6250. They even got us a second drive and
that didn't seem to cure the problem. The problem at this point is
stabilized but there isn't all that much confidence that the tape
drive is all that reliable all the time.
Have plenty of air conditioning. We kept our machine in a
temporary room which did not have that. It got moved a week and a
half ago where it has more than enough A/C. The streaming problem
seems temperature related. There was also a problem of the machine
crashing regularly at around 5:00. Temperature in the machine room
was I think 92 degrees. Very high for a machine room, but if the
machine were rated with MIL specs it still should have been running.
Oklahoma has full source, except for the kernel which they
are still waiting for. They even have the annex code.
To get the annex code they had to promise a project to Encore.
The project was to get the screen manipulation to interface with the
annex like the way the EMACS editor does. This has the net effect
of reducing the amount of work the CPU does and has it performed by the
annex.
Why we bought it:
Encore literally made a deal that was hard to beat.
They gave a package deal for the Local Area Network and the
computer. The configuration was 10 CPUs and I think we had 5 or 6
Eagle drives. Now that Encore is getting more popular I think they may
be sticking to their guns on the price. Sequent was able to match
the price for the CPUs but I think it was felt there wasn't going to
be the expandibility with the Sequent. Encore is supposed to come
out with upgrades on the Motorola chips for our machine
but I don't think OU is going to be able to afford it in the near
future. They may get them eventually but not now. Additionally,
we were allowed to log on to computers from both companies before
we bought them. We made up some benchmarks, and it appeared that
Encore seemed to come out ahead Sequnet in the floating point mark,
everything else was pretty much comparable.
You may wish to call John Hawley at (405)325-5370.
If you have any other questions, feel free to call me
or send me mail. Also, could you please send me some of the
responses you get as well. I'd like to know what others think
of these machines as well. Thanks.
Michael J. North, Systems Programmer
Cornell University Theory Center, Advanced Computing Facility
UUCP: {allegra,decvax,vax135}!cornell!batcomputer!north
Arpa: north@batcomputer.tn.cornell.edu.ARPA
ATT: (607)255-8686
Mail: 265 Olin Hall
Cornell Theory Center
Ithaca, NY 14853-5201
-------- north@tcgould.tn.cornell.edu (Michael J. North)
You may wish to ask both companies for accounts on their machines.
They both allowed us accounts and let us see what they had.
Mike North
---------- Steven M. Miller <ihnp4!umnd-cs!umnd-cs-gw!steve>
We have an Encore and for the most part have been very impressed.
They do everything they say will, give good support, but perhaps
make too many promises. They are still a small young company, and
sometimes promised software is delivered a couple of months later than
scheduled, especially third party software.
We chose an 8-processor configuration and even with 64 users on the load
average never gets about 10. Really fantastic. That's not a per
processor load, but the current number of processes that want or have
a processor! Vaxen with half the Users roll over and die with load averages
of 20 or higher in these situations. If you've seen encores simulated
load benchmark, believe it - it's true.
The current machine's biggest weakness is I/O. The machine has a fast bus,
but the processors just can't survive a real I/O load. Encore will soon
have a new cpu board that will cure the problem, so if you can wait until that
new board is available you'll be glad you did, but if really fast I/O
isn't a need then the current processors will do the job.
Perhaps the biggest problem we have with the Encore is with the networking
software. Telnetd is a real cpu pig, and since we use a lot of Bridge
equipment that creates a bit of a problem. If you plan on using
exclusively annexes, you won't have a problem. The other problem
we run into is that the retran time on Ethernet packets is about 1/3
the time of our other computers. This can really clog the net if you
have a lot of machines, or if you have a slow gateway to another network.
If you're looking to beat large loads on a vax or equivalent the Encore
is definately the way to go. The O.S. is real 4.2 except for a big
botch by Encore (in my opinion) - they threw out v7 a.out and replaced
it with COFF a.out. Since they offer a System V port and a 4.2 port this
gives them binary compatibility, but screws over any 4.2 utilities that
utilize the a.out format.
If you have any specific questions I'll do my best to answer them.
-Steve
---------- rutgers!harvard!bu-cs!bu-cs.BU.EDU!bzs (Barry Shein)
Boston University has now purchased 4 Encore systems. Two are being
used in VLSI research and the other two are to become the main focus
for computer science and engineering undergraduate coursework (the
computer science system is already well in place.)
All of our systems are roughly equivalent (they had a "university
discount" packaged system, maybe they still do) with 6 CPUS, 12MB
of memory, two ~500MB disk drives, ethernet, 4.2bsd, two Annex boxes
(16 terminal port ethernet switches), umm, oh yeah, 6250 tape
drives (one each), that's about it.
We're happy. Happy with the machines, happy with the company and,
particularly, happy with the prospects for the future such as the
new CPU boards which roughly triples performance. When they're
available we'll probably start upgrading other things like more
memory and probably more CPUs.
Even now the 6 CPU model in CS is hardly ever loaded, we'll see
around 20 users typically most of the afternoon-evening and one
can reasonably assume that the students are very active. More
active than faculty anyhow who log in from their office and then
go idle for hours (students usually use public terminal clusters,
not much motivation to log in and then go teach or something.)
The software is good 4.2bsd, networking etc works fine. Any bugs or
other problems we've had are continually being fixed, obviously any
multi-processor UNIX is a new thing and will have its own peculiar
bugs but this has not interfered with operations and new releases seem
to be getting much better (I guess the hard part is over for them, the
system basically works well.) There are a few differences such as no
/dev/kmem hacking (if you think about it it's not a natural construct
for an MP system) but they have a sys subr which provides a lot of the
same information. There are new system software facilities for
multi-processing (shared memory, locks, semaphores, threads.) It's all
pretty exciting (gee, wish I had more time to exploit it!)
If you have any more specific questions feel free to send mail.
The bottom line is I recommend Encores. I haven't any experience
with the other boxes to disrecommend them.
-Barry Shein, Boston University
-------------
4. OTHERS
-------------
From: Steven M. Miller <ihnp4!umnd-cs!umnd-cs-gw!steve>
John Mangrich <grich%max.uci.edu@ics.uci.edu>
Steven send me a survey made by John Mangrich of UC Irvine.
This is an excellent survey. Thanks, Steven and John.
Those need a copy of John's survey please send a mail to me of
them.