mo@seismo.CSS.GOV (Mike O'Dell) (10/19/87)
Greetings, I am looking for recommendations based on real experience regarding VME-based 68020 boards which are to be used in a multiprocessor system. Further, since I havd been told that the choice of backplane can make as much difference as the cpu choice, please include that in in your comments. As for how many is "multi" - a minimum of three to say at most six (that is now many I can forsee getting funding to buy!!). Further, recommendations for VME/SCSI controllers is also welcomed. I view SCSI controllers the same way I view Ethernet controllers - they should be simple but very fast. They should be just smart enough to do full duplex DMA, but in general, should NOT have a CPU on board to slow things down. -Mike O'Dell
grenley@nsc.nsc.com (George Grenley) (10/20/87)
Warning, this might be construed as a product plug, but then again, you may also find it useful... In article <44165@beno.seismo.CSS.GOV> mo@seismo.CSS.GOV (Mike O'Dell) writes: >I am looking for recommendations based on real experience >regarding VME-based 68020 boards which are to be >used in a multiprocessor system. May I recommend an alternate processor? If your intended application is CPU intensive (which I assume it is since you want multiprocessing), you will probably be interested in very high performance CPUs. My kind, generous and wise employer (it's coming up on salary review time, folks) makes a VME cpu board based on our 32532 cpu chip. The '532 is very powerful, about 9600 dhrystones at 20 mhz on our VME board. The board has the cpu, fpu, 2 sio, and 4-16meg of local ram (dual ported). It also has inter- processor interrupts, and each cpu has a user selectable ID number (1-16) to make multi processor systems easy. The CPU itself has on chip MMU and on chip cache. We don't give em away ($9900) but it IS the hotrod VME board available on the market now. BTW, the reason I know this is I'm the project leader for this board - that's right, folks, a real engineer, not a marketing flack. So, call your NSC sales person, etc. >Further, recommendations for VME/SCSI controllers is >also welcomed. I'm real fond of Ciprico's stuff. We've used it with no problems. Interphase, on the other hand, is junk - their SCSI in particular.
john@datacube.UUCP (10/22/87)
>Greetings, >... Further, since I have >been told that the choice of backplane can make as much >difference as the cpu choice, please include that in >in your comments. >... > -Mike O'Dell I don't think it is so much the backplane, but the type of bus arbiter used. The VMEbus spec (rev C.1) describes three types of arbitration. They are "prioritized", "round-robin", and "single level". It allows almost any type of arbitration. You want to read and understand chapter 3 of the spec "Data Transfer Bus Arbitration". The VMEbus spec can be obtained from Motorola. I am interested in this area, but do not yet have the depth of understanding I would like. We do not (yet) make a product that has an arbiter on it. The arbiter goes in slot one of the VMEbus. A number of other functions must also be in slot one. Most single board CPUs have the slot one stuff on them as it is required for any system. Typically, because there is not a lot of real estate left after putting the CPU on the card, the arbiter implemented is low performance, or one that will tend to give lots of bus time to itself. Very likely, to get the best bus arbiting performance, you will have to buy a separate card that is a high performance slot one, or system controller board. I would love to have others respond with the merits of bus arbiting schemes they have used or designed for the VMEbus. --John Bloomfield-- Datacube Inc. Hardware Design Group 4 Dearborn Rd. Peabody, Ma 01960 VOICE: 617-535-6644; FAX: (617) 535-5643; TWX: (710) 347-0125 UUCP: john@datacube.COM, rutgers!datacube!john, ihnp4!datacube!john {cbosgd,cuae2,mit-eddie}!mirror!datacube!john
barr@convex.UUCP (11/06/87)
As far as choice of cpu and controller, you really need to eval- uate the alternatives based upon the features and performance required by your particular application. I was rather surprised to hear the extreme bias towards Ciprico over Interphase in a previous posting. My experience with both companies' SMD disk products has been good, neither product appears to be "junk". I'd like to hear grenley@nsc justify that statement with some performance or reliability numbers. The choice of backplane is very important. Be aware that there is a problem inherent to VMEbus systems that manifests itself as a noise hit on the bus arbitration lines, and it is most severe for 32-bit operation. See the September `86 issue of DIGITAL DESIGN, "Switching Transients in Microcomputer System Buses", and the April 30 '87 issue of EDN, "Simple Solution Cures Glitches on High-Speed Buses" for descriptions of the problem and some recommended solutions. Backplanes are available that isolate the A, B, and C signal rows each in it's own layer with ground plane in between, and also run ground between each signal trace in a particular layer. Don't ignore the electrical characteristics of the backplane, especially in a high-performance application.