ciaraldi@rochester.ARPA (10/10/86)
From: Mike Ciaraldi <ciaraldi> I asked about this a year ago and got some good ideas, so here goes again... I know someone with the following system: PDP-11/23 CPU 256K RAM board 2 DLV-11J 4-port serial boards Paralel printer board RLV-12 disk controller with 4 RL02's 5-inch backplane supporting 18-bit addressing. All running TSX (which we run by first booting RT-11). This system currently supports five terminals, but cannot really run with more than 3 simultaneous users. He'd like to expand the system for more users and higher speed. We may want to do this in stages. From what I've heard, we should be able to go to more memory with our 11/23 (not an 11/23 Plus). The original (1981) DEC manual says it's limited to 256K, but the current one says it can go to 4 Meg. However, we would have to either replace or modify the backplane to allow 22-bit addresses to be sent to the boards. Also, our present memory board would not work, as it only looks at 18-bit addresses. The disk controller will work OK with 22 bit addresses. Does this sound right? From a software point-of-view, what do we have to do? I had to redo the TSX sysgen to enable the additional terminals last year. I don't see where TSX is told how much memory you have. Does it figure it out automatically? Do we have to redo the RT-11 Sysgen for the extra memory? Is that hard? (never did one for RT-1 before). For the future, we may want to go to a faster processor, e.g. an 11/53 or 11/73. Do we have to worry about this when we buy our new meory board? I know PDP-11's adjust to the speed of memory automatically, but I wonder if we have to specify especially high-speed memory or something to ensure we don't have a bottleneck in the future. If we have to buy new memory to expand anyway, we miht as well buy some that would work with a new processor too. Again, I don't see where TSX cares what model processor we have. Are there any differences? (e.g. memory managment?) Again, do we have to redo the TSGEN and/or the RT-11 sysgen when we change processors? Finally, how many users could we realistically expect to run on an 11/73? They would mostly be running accounting software, no eiting or compiling. As I said, the 11/23 with 256K poops out at 3, mainly due to lack of memory to run the jobs. I appreciate any ansswers, commets, or suggstions. Incidentally, theey may also bite the bullet and buy a Microvax II instead of upgradingthe present system. Mike Ciaraldi ARPA: ciaraldi@rochester Usenet: seismo!rochester!ciaraldi
cetron@utah-cs.UUCP (Edward J Cetron) (10/13/86)
1. go to the 11/73 (bugs still in the 11/53) 2. if the backplane is not VERY old, string the extra four lines is very simple. 3. dump the dlv11-j's since A. they only do single character i/o B. they are only good at 1200 baud sustained C. they improperly pass bus signals D. generally are disgusting (direct quote from a DEC hardware developer) 4. add more memory and a dhv11 (or third party equivalent) the above are objective expansions...the 5th suggestions is from a personal point of view: 5. upgrade to RSX-11M-Plus to take advantage of all the new hardware (i/d space and supervisor mode.....as well as a MUCH much better terminal driver...) -ed cetron Center for Engineering Design 3168 MEB, Univ. of Utah
cetron@utah-cs.UUCP (Edward J Cetron) (10/18/86)
[sacrificial line....] I sent the original article as well as my response on to a friend within digital, his comments are indented in [] ------------------------------------------------------------------------ From: Mike Ciaraldi <ciaraldi> I asked about this a year ago and got some good ideas, so here goes again... I know someone with the following system: ..... From what I've heard, we should be able to go to more memory with our 11/23 (not an 11/23 Plus). The original (1981) DEC manual says it's limited to 256K, but the current one says it can go to 4 Meg. However, we would have to either replace or modify the backplane to allow 22-bit addresses to be sent to the boards. Also, our present memory board would not work, as it only looks at 18-bit addresses. The disk controller will work OK with 22 bit addresses. [ Pitch the whole mess and start over. The backplane upgrade may or may not be easy, the 23 (possibly with some etch changes) may support 22 bit addresses, some of the older 18 bit options used BDAL<21:18> as test signals, the 18 bit memories won't work and the resulting configuration isn't maintainable. Besides, an 11/53 would be cheaper than the upgrade work. ] .... I don't see where TSX is told how much memory you have. Does it figure it out automatically? Do we have to redo the RT-11 Sysgen for the extra memory? Is that hard? (never did one for RT-1 before). [ RT-11 doesn't really take advantage of more memory, except for limited memory management control in the XM monitor. I don't know all that much about it, but I believe that RT restricts code to running in the first 28K of memory and allows data manipulation in the rest. TSX may take care of this in their timeshare mechanism, but I'd look carefully at it first. ] Finally, how many users could we realistically expect to run on an 11/73? They would mostly be running accounting software, no eiting or compiling. As I said, the 11/23 with 256K poops out at 3, mainly due to lack of memory to run the jobs. { 11/53 should run 7 - 10, 11/73 should cover 8 - 14 ejc} I appreciate any ansswers, commets, or suggstions. Incidentally, theey may also bite the bullet and buy a Microvax II instead of upgradingthe present system. [ Best idea I've heard yet. ] { and now comments on MY previously composed response - ecj} 1. go to the 11/73 (bugs still in the 11/53) [ The 11/53 is just FINE, thank you ] 2. if the backplane is not VERY old, string the extra four lines is very simple. ... 5. upgrade to RSX-11M-Plus to take advantage of all the new hardware (i/d space and supervisor mode.....as well as a MUCH much better terminal driver...) [ If they're going to convert an interactive accounting application, may as well go to VMS. ] { I then received a query , from someone at Digital re: the 11/53 bugs so I asked my same contact about the 11/53....} I recall several problems with the 11/53 that was to go on the display floor....I can't recall exactly, but I remember lots of bitching about it....it seem to recall that the standard configuration for the 11/53 barely has enough memory to do anything constructive (much less also run the 3 billion different demo programs required in the exhibit hall....) I will check with my Dec contacts and try to clarify and get you more information.... -ed [ Personally, I love the 11/53. However, that particular 11/53 had two problems: 1) Nobody brought the add on memory for the WPS+ demo or normal operation, and: 2) Nobody brought the TK50 drive I was promised, and the easiest way to install the demos is from this TK50 I have with a bootable BRUSYS and the system image. This meant copying 30+ floppies on the show floor. ...... The only complaint I have with the 53 is that it slows down when accessing off board memory, although it's still pretty quick. As for design reliability, we never breadboarded the 53 CPU, we went straight to etch, and the rev 1 etch produced workable processors. I think it was etched only two or three times before ship (as opposed to the mucho many for other modules) and I think it's one of the cleanest most reliable designs we've ever done. Some of the other complaints you may have heard from me were over a minor difference in the clock CSR which plays havoc with system software until you know about it. The KW11-L ready bit is read only. ----------------------------- { sorry for the long confusing article, but I wanted to get more information out as well as correct the references to the 11/53.... ejc } -ed cetron