jbreeden@netcom.uucp (John Breeden) (07/21/90)
Does anyone know any details about AT&T's Starserver E? The lit I have says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). They claim a MIP rating of 106MIPS (yea!, I know. That's faster than a 3090!), and they are shipping it!
davidsen@antarctica.crd.GE.COM (william E Davidsen) (07/23/90)
In article <12487@netcom.UUCP>, jbreeden@netcom.uucp (John Breeden) writes: |> Does anyone know any details about AT&T's Starserver E? The lit I have |> says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). |> They claim a MIP rating of 106MIPS (yea!, I know. That's faster than |> a 3090!), and they are shipping it! That sounds reasonable, given the numbers Corrolary was getting from a 4 CPU Z1000, and 4 CPU 4x486@25MHz. If you have a nice load of small jobs which eat the cpu and high load average, then you should benefit. If your load average is low (1 cpu hog) then this won't help much. As a guess, if vmstat shown more than one runnable process most of the time (not the same as load average) and/or the system cpu time is high, you will get a boost from the extra cpu's. I did some benchmarks (informal) for troff server use, using something like: soelim mytest.n | tbl | equ | troff -p | roff2ps > output.ps and the results were impressive. Note that I have five processes in the pipe to insure making the multicpu systems look good. The Corrolary really did! -- Bill Davidsen (davidsen@crdos1.crd.ge.com, uunet!crdgw1!crdos1!davidsen) GE Corp R&D Center, Schenectady NY Moderator of comp.binaries.ibm.pc and 386users mailing list "Stupidity, like virtue, is it's own reward" -me
rubin@cbnewsl.att.com (Mike Rubin) (07/23/90)
In article <12487@netcom.UUCP> jbreeden@netcom.uucp (John Breeden) writes: >Does anyone know any details about AT&T's Starserver E? The lit I have >says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). >They claim a MIP rating of 106MIPS (yea!, I know. That's faster than >a 3090!), and they are shipping it! Neither the additional processor boards, nor an operating system to handle them, exist yet. The box does contain the inter-processor bus so you will be able to just plug in the extra processors. The 106-MIP rating is calculated, not observed. (It's my opinion that AT&T public relations is being less than clear on these points.) What you can buy today is the 1 processor configuration, which is a full-out 33 MHz 486 with Weitek 4167 math coprocessor and DPT EISA-bus SCSI disk controller. Production is still quite limited due to the scarcity of 33 MHz 486's. I wouldn't doubt the 26 MIPS single-processor figure; it really does scream. SCSI performance is in line with other DPT controllers (not in line with a 3090 :-), more and faster disks might help. Unix (SVR4.0) runs on the machine but the Unix C compiler does not yet support the Weitek. The parallel operating system is being jointly developed with Pyramid and bears some resemblance to what you see on current Pyramid machines (except that it's SVR4.0 based instead of BSD). The "true symmetrical multiprocessor" stuff means that (unlike e.g. the dual-processor Compaq) all processors will be able to run kernel code and do I/O. --Mike Rubin <mike@attunix.att.com, leaving soon for heaven.knows.where> [Disclaimer: So far as I know everything stated here is public knowledge.]
aland@infmx.UUCP (Colonel Panic) (07/24/90)
In article <12487@netcom.UUCP> jbreeden@netcom.uucp (John Breeden) writes: >Does anyone know any details about AT&T's Starserver E? The lit I have >says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). >They claim a MIP rating of 106MIPS (yea!, I know. That's faster than >a 3090!), and they are shipping it! ^^^^^^^^^^^^^^^^^^^^ to customers? really? I had been told that the multiprocessor 486s required SVR4.1 (and, the literature I have says this), and we can't even get SVR4.0 yet for our 6386s through conventional channels... Keep in mind that MIPS comparisons to mainframes are meaningless for at least two reasons. One, the 370 instruction set is MUCH larger than a 486's, so it can do more with each instruction (e.g. conversion from packed decimal to binary is ONE assembler instruction!) Two, they use distributed I/O, so the CPU doesn't have to worry about losing cycles to disk or terminal I/O. Mainframe vs. mini/micro religious war flames to /dev/null... -- Alan Denney @ Informix Software, Inc. "We're homeward bound aland@informix.com {pyramid|uunet}!infmx!aland ('tis a damn fine sound!) ----------------------------------------------- with a good ship, taut & free Disclaimer: These opinions are mine alone. We don't give a damn, If I am caught or killed, the secretary when we drink our rum will disavow any knowledge of my actions. with the girls of old Maui."
reisert@ricks.enet.dec.com (Jim Reisert) (07/24/90)
In article <12487@netcom.UUCP> jbreeden@netcom.uucp (John Breeden) writes: > >Does anyone know any details about AT&T's Starserver E? The lit I have >says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). >They claim a MIP rating of 106MIPS (yea!, I know. That's faster than >a 3090!), and they are shipping it! Then, in a followup, <1990Jul23.165921.28420@cbnewsl.att.com>, rubin@cbnewsl.att.com (Mike Rubin) writes... > >I wouldn't doubt the 26 MIPS single-processor figure; it really does scream. >... >The 106-MIP rating is calculated, not observed. It seems hard for my tiny brain to fathom that the quad processor system runs *more than* 4 times faster than 4 individual processors (4x26=104 MIPS). If this is indeed true, we'll be seeing some interesting patents coming out of AT&T. jim =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= "The opinions expressed here in no way represent the views of Digital Equipment Corporation." James J. Reisert Internet: reisert@ricks.enet.dec.com Digital Equipment Corp. UUCP: ...decwrl!ricks.enet!reisert 77 Reed Road Hudson, MA 01749-2895
alanf@daysinns.UUCP (Daniel Alan Fleming) (07/24/90)
In article <12487@netcom.UUCP> jbreeden@netcom.uucp (John Breeden) writes: >Does anyone know any details about AT&T's Starserver E? The lit I have >says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). >They claim a MIP rating of 106MIPS (yea!, I know. That's faster than >a 3090!), and they are shipping it! Well, I don't know what details you are looking for. I have looked over the literature as well and the stats look impressive. While I was at Comdex '90 I dropped by the AT&T booth and looked over the Starserver. It is apparently being released with a number of different configurations with 1 to 4 486 chips. It is running Unix SysV 4.0. (I believe this is the first AT&T machine to be sold with it.) They claimed it could handle up to 200 PC's on Starlan 10 or 75 users interactively. I looked good but is larger than the 6386E PC server. I understand it was released at the same time as AT&T Rhapsody their windows/new wave environment and was not mentioned in its brochures. The prices I heard seem pretty amazing for the power, but a little much for a PC server, even one this powerful. I imagine it is based strickly at large companies or universities. (Who else needs that kind of power?) Hope this did some good. - Alan ========================================================================= Daniel Alan Fleming {uunet}!gatech.edu!daysinns!alanf 4203 Buford Hwy NE Days Inns of America Apt B-7 2751 Buford Hwy N.E. Atlanta, GA 30345: (404)/634-8014 Atlanta, GA 30324: (404)/728-4498 ========================================================================= Newsgroups: comp.sys.ibm.pc.hardware,comp.sys.att Subject: Re: 106 mip PC References: <12487@netcom.UUCP> Reply-To: alanf@daysinns.UUCP (Daniel Alan Fleming) Organization: Days Inns of America In article <12487@netcom.UUCP> jbreeden@netcom.uucp (John Breeden) writes: >Does anyone know any details about AT&T's Starserver E? The lit I have >says it's a true symmetrical, multiprocessor 33mhz 486 (4 processors). >They claim a MIP rating of 106MIPS (yea!, I know. That's faster than >a 3090!), and they are shipping it! Well, I don't know what details you are looking for. I have looked over the literature as well and the stats look impressive. While I was at Comdex '90 I dropped by the AT&T booth and looked over the Starserver. It is apparently being released with a number of different configurations with 1 to 4 486 chips. It is running Unix SysV 4.0. (I believe this is the first AT&T machine to be sold with it.) They claimed it could handle up to 200 PC's on Starlan 10 or 75 users interactively. I looked good but is larger than the 6386E PC server. I understand it was released at the same time as AT&T Rhapsody their windows/new wave environment and was not mentioned in its brochures. The prices I heard seem pretty amazing for the power, but a little much for a PC server, even one this powerful. I imagine it is based strickly at large companies or universities. (Who else needs that kind of power?) Hope this did some good. - Alan ========================================================================= Daniel Alan Fleming {uunet}!gatech.edu!daysinns!alanf 4203 Buford Hwy NE Days Inns of America Apt B-7 2751 Buford Hwy N.E. Atlanta, GA 30345: (404)/634-8014 Atlanta, GA 30324: (404)/728-4498 =========================================================================
cjc@ulysses.att.com (Chris Calabrese[mav]) (07/24/90)
In article <4860@infmx.UUCP>, aland@infmx.UUCP (Colonel Panic) writes: > Keep in mind that MIPS comparisons to mainframes are meaningless for > at least two reasons. One, the 370 instruction set is MUCH larger > than a 486's, so it can do more with each instruction (e.g. conversion > from packed decimal to binary is ONE assembler instruction!) > Two, they use distributed I/O, so the CPU doesn't have to worry > about losing cycles to disk or terminal I/O. The second one is _definitely_ true, but the first is more questionable. While the 370 may be able to do packed decimal to binary in one assembler instruction, there's no guarantee that it can actually execute faster than a tightly coded assembler routine for the 486. I've never used a 370, but on the VAXen we have around here, we've written assembler routines that run faster than some of the micro-coded instructions (the trick is to get all the code and data to fit in the cache). Remember, one instruction != 1 cycle (except on certain RISC architectures). MIPS ratings, in theory, are calculated by taking the number of cycles that would be needed to run some algorithm and deviding by a magic constant (like the number of cycles it would take on a VAX, hence VAX equivelant MIPS), finally doing the calculations with clock speed, memory refresh times, number of cycles for a memory read or write, etc. This is _supposed_ to get equivelant numbers for equivelant CPU horsepower. On the other hand, CPU horsepower is usually just a small fraction of what's going on in most mainframe applications. Name: Christopher J. Calabrese Brain loaned to: AT&T Bell Laboratories, Murray Hill, NJ att!ulysses!cjc cjc@ulysses.att.com Obligatory Quote: ``pher - gr. vb. to schlep. phospher - to schlep light.philosopher - to schlep thoughts.''