fdl@cbnewsh.ATT.COM (f.d.la rocca) (06/29/89)
I am trying to gain an understanding of the kind of performance Multibus II and VME backplanes actually deliver. The claim is that MBII and VME get about 32Mbps. However, when measured thru-put is quite a bit less because of the software overhead in the case of MBII. Not being as familiar with VME I don't think software over head is as big and issue in that its shared memory vs. message passing. In discussions I have had with others about this subject a better metric would be to compare transactions per second from say a host board to a network board over VME or MBII. This would be an apples to apples type comparison. What I am looking for are papers or any measured data available. Has anyone measured the transaction per seconds accross MBII or VME? What kind of real performance do hi-performance backplanes deliver other than MBII or VME that are on the market today? What do machines that are targeted at network server applications typically achieve.
ron@motmpl.UUCP (Ron Widell) (07/03/89)
In article <1796@cbnewsh.ATT.COM> fdl@cbnewsh.ATT.COM (f.d.la rocca) writes: >I am trying to gain an understanding of the kind of >performance Multibus II and VME backplanes actually deliver. >The claim is that MBII and VME get about 32Mbps. ^^^^^^ Do you really mean 32 Mega-bits-per-second? >However, when measured thru-put is quite a bit less >because of the software overhead in the case of MBII. Not being >as familiar with VME I don't think software over head is as big and issue >in that its shared memory vs. message passing. [stuff deleted] Well it's been quite a while since I've looked at either of the specs (and I'm at home right now) (WARNING-possible memory parity error) but it seems to me that the theoretical max for MBII was about 40 MBytes/sec (one 4-byte transfer every 100nS) while the theoretical max for VME was about 48 Mbytes- per-second (one 4-byte transfer every 60 nS). Neither of them ever (to my knowledge) got that. But both of them, given the proper memory system, could comfortably run in the 20-30 Mbyte/sec range. I have personally measured > 100Mbits/sec on a VME backplane. (2 50Mbit/sec serial channels plus transaction overhead). I am really out of touch with the NuBUS and FutureBus activity so someone else will have to provide some guidance there. Regards, -- Ron Widell, Field Applications Eng. |UUCP: {...}mcdchg!motmpl!ron Motorola Semiconductor Products, Inc., |Voice:(612)941-6800 9600 W. 76th St., Suite G | I'm from Silicon Tundra, Eden Prairie, Mn. 55344 -3718 | what could I know?
smith@iwblsys.UUCP (Mickey Smith) (07/06/89)
In article <1284@motmpl.UUCP>, ron@motmpl.UUCP (Ron Widell) writes: > > Well it's been quite a while since I've looked at either of the specs (and > I'm at home right now) (WARNING-possible memory parity error) but it seems > to me that the theoretical max for MBII was about 40 MBytes/sec (one 4-byte > transfer every 100nS) while the theoretical max for VME was about 48 Mbytes- > per-second (one 4-byte transfer every 60 nS). > Neither of them ever (to my knowledge) got that. But both of them, given the > proper memory system, could comfortably run in the 20-30 Mbyte/sec range. MultiBus II backplane speed is based on a 10 MHz synchronous clock and a 32 bit data path. MultiBus II will only make transfers across the bus at this rate(40 Megabytes/sec), no faster and no slower. As far as bus performance, consider this. The transfer rate from one memory subsystem to another is dictated by the slowest memory subsystem speed. This is true of VME as well as Multibus II when we compare such memory to memory data movements. But since there is a bus interface part on Multibus II (the Message Passing Coprocessor) that decouples the local bus from the backplane bus, the time spent holding the backplane for the transfer is held to a percent of total bus bandwidth. To illistrate this: Time to transfer from memory A to memory B: |----------------------------------------------------------| Packetized transfers on the backplane bus: |--| |--| |--| |--| |--| |--| The time between the packetized transfers is usable for other agents on the bus to carry out other functions. -- Mickey Smith, Regional Software Specialist | Standard disclaimer, I speak Intel Corporation | only for myself 7071 Orchard Lake Road | W. Bloomfield, MI 48332 (313) 851-8096 |
woods@b8.UUCP (Mike Wood) (07/11/89)
If anyone can help me, I'm searching for a text book that covers the VME arch. and theory of operation. My company offers what we call a VME adapter interface board. I have yet to find any information on the VME. Any assistance would be greatly appreciated. Thank You In Advance, Mike Wood, Intergraph Huntsville, Alabama