YEHAVI@vms.huji.ac.il (02/02/90)
I have a question which I am very curious to know the answer for:
Take a VAX-6300 which is rated at 4MIPS and a RISC machine which is rated
as over 14 MIPS. THe VAX 6300 can hold over 100 users (few are doing heavy
jobs, most do light jobs) and a few batches. The RISC machine from the other
side is rated by its manufacturer to 35 users... Altough there are a lot
types of RISC machines I hear from most suppliers that RISC machine cannot
hold a large number of users, but very few suppliers claim they can.
So, what is the problem with the RISC machines? Context switching,
initiating a lot of I/O requests, or what? (I am talking on a mixed academic
environement which includes compilations, statistics packages, databases,
etc., little from everyhting).
Thanks in advance,
__Yehavi:
Yehavi Bourvine
The Hebrew University of Jerusalem
YEHAVI@HUJIVMSmark@mips.COM (Mark G. Johnson) (02/02/90)
Three keys on my keyboard are busted, the ones to the right of
shift-L, 0, and shift-9. So take "L0(" and shift right one key.
>> Altough there are a lot types of RISC machines I hear from most
>> suppliers that RISC machine cannot hold a large number of users,
>> but very few suppliers claim they can. So, what is the problem
>> with the RISC machines? Context switching,
The problem with Reduced Instruction Set Computers is that the
instruction set determines how many users a computer can handle.
Some instruction sets (for example, the VAX instruction set) can
handle a lot of users, while others (like, the i80286 instruction
set) cannot. Reduced Instruction Set Computers ("RISC") just don't
have enough instructions to adequately take care of a large
set of simultaneous users. L0(
Think about it; some people say that the Crays were the first RISC
machines, yet Crays are just batch scheduled compute engines; they
handle effectively ONE user (!).
You really need to have about 5 or 6 instructions per user, so a
RISC with its 100-120 instructions tops out at maybe 20-30 users.
{and have you noticed that they count the different options of
branches as different instructions?: branch-on-less-than and
branch-on-greater-than are counted as two instructions. Seems a
little bit like cheating, doesn't it?} On the other hand, a VAX
has 500 instructions --- and, best of all, it has microcode and
microinstructions, which count for more. L0(
The VAX executes a superset of the predecessor machine's (PDP-11's)
instructions. In fact the X in VAX stands for eXtension. So
the VAX has more instructions than the PDP-11 and it can consequently
handle more users than the PDP-11 did.
Similarly, the IBM 370 instruction set is a superset of the IBM 360.
And the 370 handles lots more users than the 360 ever did.
Summary: (# of users) = (1/5) * (# of instructions).
This is why RISCs will always be inferior to the VAX and
the S/370 in multiuser capability.
This fundamental truth has been kept secret for a long time,
by the same devilish worldwide conspiracy that first put fluoride
in drinking water.
A homework problem: Two VAX machines are the microVAX and the
VAXstation. Compare these to the Sun 4/490 computer [SPARC "RISC"
instruction set] and the MIPS M/2000 computer [MIPS "RISC" instruction
set] in terms of number of simultaneous users. Draw a conclusion
about # of users vs. RISC/VAX instruction sets.
Signing off with a right shift of one keyboard position "L0(" .....
--
-- Mark Johnson
MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086
(408) 991-0208 mark@mips.com {or ...!decwrl!mips!mark}neil@uninet.cpd.com (Neil Gorsuch) (02/03/90)
In article <25c936b0141@vms.huji.ac.il> YEHAVI@vms.huji.ac.il writes: > I have a question which I am very curious to know the answer for: >Take a VAX-6300 which is rated at 4MIPS and a RISC machine which is rated >as over 14 MIPS. THe VAX 6300 can hold over 100 users (few are doing heavy >jobs, most do light jobs) and a few batches. The RISC machine from the other >side is rated by its manufacturer to 35 users... Altough there are a lot >types of RISC machines I hear from most suppliers that RISC machine cannot >hold a large number of users, but very few suppliers claim they can. > So, what is the problem with the RISC machines? Context switching, >initiating a lot of I/O requests, or what? (I am talking on a mixed academic >environement which includes compilations, statistics packages, databases, >etc., little from everyhting). There are a lot of reasons, but remember that the ratings are heavily slanted towards computer manufacturers selling minicomputers and servers rather than desktop workstations. If it were just CISC versus RISC, the Microvax III would be rated at 50 to 100 users according to it's MIPS rating. 1. A lot of it is the infamous $50,000 card cage syndrome. Computer manufacturers have to very competitive with their desktop workstations that have a great MIPS rating, but they always seem to charge a large amount for card cages. 2. There is a certain amount of problems with context switching. I don't know about Dec RISC machines, but in the case of Suns, a SPARC processor has a fixed number of register windows, which are very easy to context switch between. But I would much rather be one of 35 people using a 12 MIPS Sparcstation with 64 Mbytes of memory, assuming that most were light usage, than one of 100 people using a 4 MIPS VAX 6300. 3. The larger machines have higher speed i/o busses, but a lot of disk time is spent waiting for the head to get to the proper spot on the disk, and you can buy 10.5 mS average seek time SCSI disks with over 1 Mbyte/second transfer rates. 4. The smaller machines do not usually have any means of adding expansion cards for things like serial ports, and putting 50 people on one ethernet leg would impact ethernet performance, but there are now cheaper ethernet serial interfaces, and even a cheaper still SCSI serial interface that won't degrade the ethernet. All in all, I would say that a Decstation (assuming that you can add enough memory) can handle at least half the users that a 6300 can. And I doubt that putting 100 people on a 4 MIPS machine is ever going to please the users of that machine, anyway. And even if it's only a third the users, the price difference is a lot more than 3 times! If you are stuck with giving people terminals, what I would do is use a Decstation for every 16 users, and if you shop around for the peripherals, you can buy a 16 user system, including Decstation, disk(s), serial interfaces and terminals for about $25,000 total 8-). And the users will be much more pleased with almost a MIP each, and think of the money you will save on service contracts 8-). Or if the users are used to noticable key stroke response delays, put 32 people on a Decstation for around $35,000. -- Neil Gorsuch INTERNET: neil@cpd.com UUCP: uunet!zardoz!neil MAIL: 1209 E. Warner, Santa Ana, CA, USA, 92705 PHONE: +1 714 546 1100 Uninet, a division of Custom Product Design, Inc. FAX: +1 714 546 3726 AKA: root, security-request, uuasc-request, postmaster, usenet, news
burzio@mmlai.UUCP (Tony Burzio) (02/05/90)
In article <35387@mips.mips.COM>, mark@mips.COM (Mark G. Johnson) writes: I hope the original poster was joking, but with all those gullible VAX architecture users out there, one can hardly take the chance :-) > The problem with Reduced Instruction Set Computers is that the > instruction set determines how many users a computer can handle. Perhaps one should also look at the bus structure and I/O bandwidth in the quest for number of users. > Some instruction sets (for example, the VAX instruction set) can > handle a lot of users, while others (like, the i80286 instruction > set) cannot. Reduced Instruction Set Computers ("RISC") just don't > have enough instructions to adequately take care of a large > set of simultaneous users. L0( Odd one that. How then, do you explain the HP computer (RISCish) that can handle 500+ users doing transaction processing? No, the number of instructions has only minor import. Perhaps what you are seeing is the well known Sun saw that says: "RISC = 15 MIPS downhill with a stiff tail wind" :-). When measuring speed, first use MIPS, then number of users. The kernal can be tuned for low number of users and get higher performance benchmark numbers. > Think about it; some people say that the Crays were the first RISC > machines, yet Crays are just batch scheduled compute engines; they > handle effectively ONE user (!). Could this be that CRAY computer memory is very expensive, and much faster than disk swapping? Forcing a CRAY to swap removes the advantage of this super-fast memory... > has 500 instructions --- and, best of all, it has microcode and > microinstructions, which count for more. L0( Hmmm... Our HP835 (~14 MIPS) handles under load much better than our VAX3xxx. The HP835 is rated 30 users actually DOING something, not like our VAXen which just sit around with 40 users idling waiting for mail. A 1 week job on a RISC machine is a 4 week job on a VAX :-) > the VAX has more instructions than the PDP-11 and it can consequently > handle more users than the PDP-11 did. Perhaps modern chip sets could help a PDP-11 blow away a VAX? But then, one could ask "Why?" > Summary: (# of users) = (1/5) * (# of instructions). > This is why RISCs will always be inferior to the VAX and > the S/370 in multiuser capability. Hmmm... Not according to Digital Review, which talked about new 500+ user RISC machines. One could load down a RISC machine so that it had the response time of a VAX, but I have found that people get RISC machines to get AWAY from slow computers like the VAX. We have found that there is a general rebellion among our users who are getting their own RISC fast machines to be done with central VAXen... The users are always right :-) ********************************************************************* Tony Burzio * Man the pumps! Martin Marietta Labs * mmlai!burzio@uunet.uu.net * *********************************************************************
brent@uwovax.uwo.ca (Brent Sterner) (02/07/90)
In article <673@mmlai.UUCP>, burzio@mmlai.UUCP (Tony Burzio) writes: > In article <35387@mips.mips.COM>, mark@mips.COM (Mark G. Johnson) writes: > > I hope the original poster was joking, but with all those gullible > VAX architecture users out there, one can hardly take the chance :-) > No, he was serious... > < stuff deleted > > > Hmmm... Not according to Digital Review, which talked about new 500+ > user RISC machines... Don't you see they cheated! They added a *lot* of CISC instructions, but just didn't *use* them. Must have added a lot of unused ucode instructions too (to get the sorts of numbers DR quotes). To the author of the DR article, I'm really disappointed! DR usually prides itself on catching cheats. Shame, shame, ohh double-shame! b. -- Brent Sterner Technical Support Manager, Academic Systems Network <BRENT@uwo.ca> <BRENT@UWOVAX.BITNET> <129.100.2.13> Telephone (519)661-2151 x6036 Last Gasp Computing & Communications Services, Natural Sciences Building The University of Western Ontario, London, Ontario, Canada N6A 5B7
marbru@auto-trol.UUCP (Martin Brunecky) (02/07/90)
In article <673@mmlai.UUCP> burzio@mmlai.UUCP (Tony Burzio) writes: >In article <35387@mips.mips.COM>, mark@mips.COM (Mark G. Johnson) writes: > >I hope the original poster was joking, but with all those gullible >VAX architecture users out there, one can hardly take the chance :-) > >> The problem with Reduced Instruction Set Computers is that the >> instruction set determines how many users a computer can handle. > ... I'v been always considered a VAX bigot, but enough is enough. But expressing number of users machine can handle in terms of the instruction set size is rudiculous. May be I don't get the joke. There are, however, differences between RISC and CICS machines that may explein why 2 times faster (MIPS count) RISC box can not always handle 2 times more users. 1) Code size. CICS stands for "complex" instruction set. Thus, to accomplish the same operation, most RISC architectures need more instructions. More instructions = more memory to hold'em, more FETCHes to get them from the memory, more I/O to page'em into memory. On busless desktop machines this ain't that important, but for most "midranges" use some kind of memory bus (to get to it's 256 or more MB), and this beast is soon prone to become a bottleneck. 2) Cache usage. The larger code size naturally requires larger instruction cache - otherwise all that the faster RISCy box can do is WAIT FASTER for the memory cycle completion. 3) Context switch impact. In multiuser environment, a context switch is the name of the game. Context switch means that most registers must be re-loaded, that a significant portion of cache will become invalid ... Now, on machines with NO cache at all (MicroVAX is close to this one), the context switch will have much less impact than on SPARC machine with large cache and large register file to save. 4) Overall architecture. Now, a computer is not only the CPU chip. It has a memory, I/O controllers, disks... In those areas, the progress is not measured by MIPS. For example, the disk I/O channel has gone from 1MB/sec in 1978 to 2-4MB/sec today, while the "MIPS" jumped from 1 to 20. Overall, I don't think there is an easy answer to a quetion "how many users". The APPLICATION is the most important component of the puzzle. And the application may be tuned to take advantage of CICS (a VAX for example), or tuned to take advantage of RISC. Many people are spending hours trying to fine tune their machines (VAXen or UNIXes, does not matter), and getting gains in 0 to 20% range. Some people are tuning applications, getting gains in 100 - 500%+ range. Therefore, I would allways try to look at the combination MACHINE+APPLICATION, and forget the RISC versus CICS issue. [ The opinions above are solely subjective, without any thoretical background in the field of computer architectures ]. -- =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=* Martin Brunecky marbru@auto-trol.COM (303) 252-2499 {...}ncar!ico!auto-trol!marbru Auto-trol Technology Corp. 12500 North Washington St., Denver, CO 80241-2404
rogerk@mips.COM (Roger B.A. Klorese) (02/08/90)
In article <730@auto-trol.UUCP> marbru@auto-trol.UUCP (Martin Brunecky) writes: >In article <673@mmlai.UUCP> burzio@mmlai.UUCP (Tony Burzio) writes: >>In article <35387@mips.mips.COM>, mark@mips.COM (Mark G. Johnson) writes: >> >>I hope the original poster was joking, but with all those gullible >>VAX architecture users out there, one can hardly take the chance :-) >> >>> The problem with Reduced Instruction Set Computers is that the >>> instruction set determines how many users a computer can handle. >> > > ... I'v been always considered a VAX bigot, but enough is enough. > But expressing number of users machine can handle in terms of the > instruction set size is rudiculous. May be I don't get the joke. The problem is you *did* get the joke. That's it. It was a *joke*, see?! -- ROGER B.A. KLORESE MIPS Computer Systems, Inc. phone: +1 408 720-2939 928 E. Arques Ave. Sunnyvale, CA 94086 rogerk@mips.COM {ames,decwrl,pyramid}!mips!rogerk "Two guys, one cart, fresh pasta... *you* figure it out." -- Suzanne Sugarbaker
Frank.Mallory@f123.n109.z1.fidonet.org (Frank Mallory) (02/08/90)
YA> Take a VAX-6300 which is rated at 4MIPS and a RISC machine which is YA> rated YA> as over 14 MIPS. THe VAX 6300 can hold over 100 users (few are doing YA> heavy YA> jobs, most do light jobs) and a few batches. The RISC machine from the YA> other YA> side is rated by its manufacturer to 35 users... Altough there are a YA> lot YA> types of RISC machines I hear from most suppliers that RISC machine YA> cannot YA> hold a large number of users, but very few suppliers claim they can. Essentially the same question was answered in this week's Digital Review. The answer is that the basis for the ratings is not the same. You have to run the same benchmark on both to determine the estimated transactions per second of each and thus the relative ability to support a user load. frank.mallory@blkcat.fidonet.org -- Frank Mallory at The Black Cat's Shack (Fidonet 1:109/401) Internet: Frank.Mallory@f123.n109.z1.fidonet.org UUCP: ...!uunet!blkcat!123!Frank.Mallory
terryl@tekcrl.LABS.TEK.COM (02/08/90)
+Summary: one keyboard position right of "L0("
+Three keys on my keyboard are busted, the ones to the right of
+shift-L, 0, and shift-9. So take "L0(" and shift right one key.
+
People, here he is telling you "It's a joke, son!!!" (in my best Fog-
horn Leghorn voice!!!). What happens when you take "L0(" and shift that right
one key??? TTAAADDDAAA!!!! :-), a smiley face; and just about everyone fell
for it......
+ >> Altough there are a lot types of RISC machines I hear from most
+ >> suppliers that RISC machine cannot hold a large number of users,
+ >> but very few suppliers claim they can. So, what is the problem
+ >> with the RISC machines? Context switching,
+
+
+The problem with Reduced Instruction Set Computers is that the
+instruction set determines how many users a computer can handle.
+Some instruction sets (for example, the VAX instruction set) can
+handle a lot of users, while others (like, the i80286 instruction
+set) cannot. Reduced Instruction Set Computers ("RISC") just don't
+have enough instructions to adequately take care of a large
+set of simultaneous users. L0(
+
+Think about it; some people say that the Crays were the first RISC
+machines, yet Crays are just batch scheduled compute engines; they
+handle effectively ONE user (!).
+
+You really need to have about 5 or 6 instructions per user, so a
+RISC with its 100-120 instructions tops out at maybe 20-30 users.
+{and have you noticed that they count the different options of
+branches as different instructions?: branch-on-less-than and
+branch-on-greater-than are counted as two instructions. Seems a
+little bit like cheating, doesn't it?} On the other hand, a VAX
+has 500 instructions --- and, best of all, it has microcode and
+microinstructions, which count for more. L0(
+
+The VAX executes a superset of the predecessor machine's (PDP-11's)
+instructions. In fact the X in VAX stands for eXtension. So
+the VAX has more instructions than the PDP-11 and it can consequently
+handle more users than the PDP-11 did.
+
+Similarly, the IBM 370 instruction set is a superset of the IBM 360.
+And the 370 handles lots more users than the 360 ever did.
+
+Summary: (# of users) = (1/5) * (# of instructions).
+ This is why RISCs will always be inferior to the VAX and
+ the S/370 in multiuser capability.
+
+This fundamental truth has been kept secret for a long time,
+by the same devilish worldwide conspiracy that first put fluoride
+in drinking water.
+
+A homework problem: Two VAX machines are the microVAX and the
+VAXstation. Compare these to the Sun 4/490 computer [SPARC "RISC"
+instruction set] and the MIPS M/2000 computer [MIPS "RISC" instruction
+set] in terms of number of simultaneous users. Draw a conclusion
+about # of users vs. RISC/VAX instruction sets.
+
+Signing off with a right shift of one keyboard position "L0(" .....riley@batcomputer.tn.cornell.edu (Daniel S. Riley) (02/08/90)
In article <455.25D0A8A9@blkcat.fidonet.org> Frank.Mallory@f123.n109.z1.fidonet.org (Frank Mallory) writes: > YA> types of RISC machines I hear from most suppliers that RISC machine > YA> cannot > YA> hold a large number of users, but very few suppliers claim they can. > >Essentially the same question was answered in this week's Digital Review. The >answer is that the basis for the ratings is not the same. You have to run the >same benchmark on both to determine the estimated transactions per second of >each and thus the relative ability to support a user load. At the Anaheim DECUS last fall, there was a talk titled "Performance of the DECstation 3100/3100S/5400", by James Seagraves of the Ultrix(tm) Engineering Group. The talk was basically a ton-and-a-half of the sorts of graphs DEC makes to determine the recommended number of users for various systems. One thing that's very obvious looking at the graphs is that the maximum number of users depends on the kind of load on the system. For CPU intensive tasks, response time vs. number of users for a DECstation is flat way past the point where a VAX 6310 has started to bog down. For I/O intensive tasks, a DECstation 3100 doesn't have much chance against a 6310 with a KDB50, but a DS5400 with a KDA50 looks pretty good. So you have to look at the I/O structure and memory bandwidth as well as the MIPS rating, taking the intended use for the system into account. (I also got the impression that DEC was somewhat more conservative assigning a maximum number of users to the DECstation 3100. In their load tests, the RISC systems don't seem to bog down as dramatically as the VAXes do, so there is much more latitude in setting the cutoff point.) -Dan Riley (riley@tcgould.tn.cornell.edu, cornell!batcomputer!riley) -Wilson Lab, Cornell University