scott_j_f...@yahoo.com (Scott Ford) writes: > Werent they developed at La Gaude ? I was there in the 90s re: http://www.garlic.com/~lynn/2014f.htm#49 Beyond the EC12
4341 was being done in Endicott, maybe thinking about (slower) 4331 that was being done in Europe (Boeblingen) on 4361 (4331 followon) http://www-03.ibm.com/ibm/history/exhibits/mainframe/mainframe_PP4331.html Date: 08/26/82 12:00:34 From: wheeler to be fair, Endicott has a faster 4341mp that they won't get to announce. POK has strapped back a 3081 to create a slowdown'ed 3083 and I expect that Endicott is now under POK's thumb, they will not be allowed to do anything more in that area ... 4341 frame was engineered to hold two CPUs and 16meg of 32k OEM chips (in case IBM tried to screw them on deliveries of IBM 64k chips). The E7 would only be a little slower than the 3083. Also it is not clear from some of the high I/O benchmark reports whether or not the 3081 technology with high I/O rates & high task switch rates (lots of cycle stealing & lower cache hit ratios) is faster than a 3033. ... snip ... other past 4300 email http://www.garlic.com/~lynn/lhwemail.html#43xx 3033 was by POK 168 group ... mapping 168 logic to 20% faster chips from FS, at the same time 3081 was being done by different group As soon as 3033 is out the door, that group starts on 3090 (in parallel with the 308x efforts). I've mentioned before cluster of (original) 4341 had more aggregate processing power than 3033, more aggregate memory than 3033, more aggregate I/O than 3033, lower cost than 3033, and much lower space & environmental footprint than 3033. At one point, head of POK getting allocation of critical 4341 manufacturing component cut in half (as way of dealing with 4341 competition). With minor tweak, 4341 channels handle datastreaming 3mbytes/sec. By the time, 3083 is coming out the door, Endicott has faster 4341. 3083 http://www-03.ibm.com/ibm/history/exhibits/mainframe/mainframe_PP3083.html 3081 http://www-03.ibm.com/ibm/history/exhibits/mainframe/mainframe_PP3081.html 3081 http://en.wikipedia.org/wiki/IBM_3081 recent slightly related 370xa/3081 folklore ... or how I got to spend 3hrs being interviewed by FBI agent (recent linkedin discussion about two bldgs crammed full of old 360&370 systems have been found) http://www.garlic.com/~lynn/2014f.html#27 Complete 360 and 370 systems found 308x channels are slow and have lots of issues: Date: 09/17/82 10:40:29 From: wheeler Talking to a GPD engineer he says that IBM has not technical talent to come out with another control unit. As an example, he said that when he joined the group out here, there were at least 10-12 people in his area alone that understood the channel interface ... he thinks that there might be one such person now in the whole GPD division ... the rest have left the company. I've heard what sounds like contigency projects on the east coast with channel development that completely bypass control units and connects directly to drives. .... I was in meetings all day yesterday, but one time I stopped by my office two people down the hall were talking about head crash on 3380 and now might be a good time to sell all your IBM stock. SJRLVM1 took head crash on customer ship level of 3380s yesterday and they replaced the HDAs last night in the box. ... Performance numbers for the 3084 seem to have some liberties. 4-way should have three times the performance interferance that a 2-way (cache invalidation signals from 3 other processors instead of one). They cheat with the 3083 versis 3081. for example, on a 158ap, running a UP generated system ... the processor runs 10% slower if the switch on the machine is in AP-mode rather than UP-mode (additional delay in each machine cycle just to listen for cache invalidation signals from the other processor ... this is w/o the other processor even executing anything generating storage alterations & cache invalidation signals). For 3083 the machine cycle invalidation listening delay was left in the machine. I've heard that the 3084 numbers are somewhat selected benchmarks that do minimal storage alterations ... extensive storage alteration programs can have disastrous effects on 3084 performance. ... I've been told that almost every control unit that has attached to a 308x has had to undergo hardware ECs ... apparently it was easier for every control unit hardware group in the company (even on machines no longer with development group people available) to resolve the problems than for the 308x channels. Also did you see the message that ACP runs 20% slower on a 3081d than on a 3033. On a 3081k, ACP runs 5% faster than a 3033. POK is started a special 3081k CPU program where the 3081s coming down the line will be tested to see if they can run with their clock cranked down. If they pass, they will be special high performance 3081Ks which run slightly faster than normal 3081ks. ... snip ... Note that there was some early 3380 quality problems with "sticktion" that required crash effort to resolve. There was problem in the 70s with lots of senior disk engineers (that understood the channel interface) leaving for startups. That was also the explanation that they gave me about increasingly insisting that I sit in on conference calls with POK channel engineers ... another one of the ways I was being ask to play disk engineer ... some past posts http://www.garlic.com/~lynn/subtopic.html#disk -- virtualization experience starting Jan1968, online at home since Mar1970 ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN