wsd@cs.brown.edu (Wm. Scott `Spot' Draves) (02/23/90)
I read in Unix Today! that the new IBM RISCs used an improved MCA bus that doubled its performance from 20Mb/s to 40Mb/s. There is a separate, higher performance, bus for memory. Isn't this quite underpowered relative to the CPU? What sort of swapping performance can I expect? For comparison, the SS1's S-bus is 100Mb/s, I think. Scott Draves Space... The Final Frontier wsd@cs.brown.edu uunet!brunix!wsd Box 2555 Brown U Prov RI 02912
Robert.Sansom@CS.CMU.EDU (02/24/90)
Excerpts from mail: 23-Feb-90 bus on new IBMs Wm. Scott `. Draves@cs.b (446) > I read in Unix Today! that the new IBM RISCs used an improved MCA bus > that doubled its performance from 20Mb/s to 40Mb/s. There is a > separate, higher performance, bus for memory. Shouldn't that be 20-40 MB/s (peak). Robert Sansom <rds@cs.cmu.edu> School of Computer Science Carnegie Mellon University
raob@mullian.ee.mu.oz.au (Richard Oxbrow) (02/24/90)
In article <WSD.90Feb22202840@pitman.cs.brown.edu> wsd@cs.brown.edu (Wm. Scott `Spot' Draves) writes: > >I read in Unix Today! that the new IBM RISCs used an improved MCA bus >that doubled its performance from 20Mb/s to 40Mb/s. There is a >separate, higher performance, bus for memory. > >Isn't this quite underpowered relative to the CPU? What sort of >swapping performance can I expect? > >For comparison, the SS1's S-bus is 100Mb/s, I think. > You really need to have a look at the specifications of both buses. In the case of the S-Bus you should read the specs. 100 Mb/s @25Mhz "burst" thru' put for the S-Bus For the case of the S-Bus on a 16 2/3 Mhz bus, a'la SS1 29 Mb/s @16 2/3Mhz "burst" thru' put (DMA) 25 Mb/s @16 2/3Mhz "burst" thru' put (DVMA) I would emphasis the the word "burst", you really need to have a close look at the S-Bus specs in relation to the architecture of the machine with the S-Bus. The S-bus however is a nice and simple bus with a reasonabley high thru' put. richard .. #ps. any hints on the next line of Suns ? Richard Oxbrow |ACSnet raob@mullian.oz dept. of ee eng ,uni of melbourne |Internet raob@mullian.ee.mu.OZ.AU parkville 3052 australia |Arpa-relay raob%mullian.oz@uunet.uu.net fax +[061][03]344 6678 |Uunet uunet!munnari!mullian!raob
ewong@gpu.utcs.utoronto.ca (Eddy Wong) (02/24/90)
In article <gZtMl3q00j8586TAoM@cs.cmu.edu> Robert.Sansom@CS.CMU.EDU writes: >Excerpts from mail: 23-Feb-90 bus on new IBMs Wm. Scott `. Draves@cs.b >(446) >> I read in Unix Today! that the new IBM RISCs used an improved MCA bus >> that doubled its performance from 20Mb/s to 40Mb/s. There is a >> separate, higher performance, bus for memory. > >Shouldn't that be 20-40 MB/s (peak). > >Robert Sansom <rds@cs.cmu.edu> >School of Computer Science >Carnegie Mellon University A 10Mhz NuBus can have a transfer rate close to 40 MB/s if block transfers are used. This compare very nicely to the improved MCA bus. Eddy Wong ewong@gpu.utcs.utoronto.ca
raob@mullian.ee.mu.oz.au (Richard Oxbrow) (02/25/90)
>>that doubled its performance from 20Mb/s to 40Mb/s. There is a >>For comparison, the SS1's Sbus is 100Mb/s, I think. >For the case of the SBus on a 16 2/3 Mhz bus, a'la SS1 > > 29 Mb/s @16 2/3Mhz "burst" thru' put (DMA) > 25 Mb/s @16 2/3Mhz "burst" thru' put (DVMA) The SS1 is actually ^^^^^^^^20 MHz , my mistake .. (thanks dc@sun.com) the rest of the numbers are correct (see. page 25, SBus Specification Rev.A) richard .. #PS. And Mb/s = mega BYTES per second (next time I will use MB/s) Richard Oxbrow |ACSnet raob@mullian.oz dept. of ee eng ,uni of melbourne |Internet raob@mullian.ee.mu.OZ.AU parkville 3052 australia |Arpa-relay raob%mullian.oz@uunet.uu.net fax +[061][03]344 6678 |Uunet uunet!munnari!mullian!raob
casey@ultra.com (Casey Cox) (02/27/90)
In article <gZtMl3q00j8586TAoM@cs.cmu.edu> Robert.Sansom@CS.CMU.EDU writes: >Excerpts from mail: 23-Feb-90 bus on new IBMs Wm. Scott `. Draves@cs.b >(446) >> I read in Unix Today! that the new IBM RISCs used an improved MCA bus >> that doubled its performance from 20Mb/s to 40Mb/s. There is a >> separate, higher performance, bus for memory. > >Shouldn't that be 20-40 MB/s (peak). > The new IBM RISCs have a new micro channel bus that supports the extended streaming protocol that allows data to be transferred every 100ns. This new 40 MB/sec is of course their peak transfer rate and I'm sure the sustained rate is somewhat lower, but it has nothing to do with the old 20 MB/sec rate on other machines without the streaming protocol. -- Casey Cox Ultra Network Technologies / 101 Daggett Drive / San Jose CA 95134 casey@ultra.com ...!ames!ultra!casey (408) 922-0100
grunwald@foobar.colorado.edu (Dirk Grunwald) (02/27/90)
at a local presentation, they said the protocol was: Level 1: --------------------------------- Addr | AAA | | AAA | | --------------------------------- Data | | DDD | | DDD | --------------------------------- ^ ^ +-------+ 100ns Data is ready every 200ns Level 2: --------------------------------- Addr | AAA | | | | --------------------------------- Data | | DDD | DDDD | DDD | --------------------------------- ^ ^ +-------+ 100ns Data is ready every 100ns, streaming only Level 3: --------------------------------- Addr | AAA | DDD | DDD | DDD | --------------------------------- Data | | DDD | DDD | DDD | --------------------------------- ^ ^ +-------+ 100ns Two data are ready every 100ns, streaming only their level 4 plan called for dropping the clock cycle to 50ns I think.
yap@me.utoronto.ca (Davin Yap) (02/27/90)
grunwald@foobar.colorado.edu (Dirk Grunwald) writes: >Level 3: > --------------------------------- > Addr | AAA | DDD | DDD | DDD | > --------------------------------- > Data | | DDD | DDD | DDD | > --------------------------------- > ^ ^ > +-------+ > 100ns > Two data are ready every 100ns, streaming only The sales rep at the announcement told me that this was their 80 MByte/s mode. True of False? --
grunwald@foobar.colorado.edu (Dirk Grunwald) (02/28/90)
>>>>> On 27 Feb 90 00:51:28 GMT, yap@me.utoronto.ca (Davin Yap) said: DY> grunwald@foobar.colorado.edu (Dirk Grunwald) writes: >Level 3: > --------------------------------- > Addr | AAA | DDD | DDD | DDD | > --------------------------------- > Data | | DDD | DDD | DDD | > --------------------------------- > ^ ^ > +-------+ > 100ns > Two data are ready every 100ns, streaming only DY> The sales rep at the announcement told me that this was their 80 MByte/s DY> mode. True of False? DY> -- could be - but, the 50ns implies 20M-DDD/second. I assume a DDD is 2 bytes? If not, and it's 4 bytes (i.e. 32bit wires), you'd get 80Mb/s, not the 40 I was thinking of. Then, by dropping to 50ns strobes, you'd get 160Mb/s. I think that sounds right. They mumbled something about their optical interface, which was supposedly 200Mb/s, but I don't think it goes through the micro-channel, although it's controlled by the same controller.