root%bostonu.csnet@csnet-relay.arpa (BostonU SysMgr) (05/23/85)
1. uvaxII It seems one of the great hopes out there is that the QDA-50 (or is it KDA-50?) controller will become available and the uvax will then support RA* series disks with their size and performance advantages. What immediately occurs to me is: a) We're talking minimum disk buy-in of about $27K for your first disk (RUA81-CA), that's a more than the system, maybe that's reasonable but worth a thought. b) More important: I think all the discussions about RA disks on this list point towards putting them into A/C machine room environments, plus they do need, I believe, L5-30Rs which you may not have in your office. I would not be surprised to hear that DEC has new RA disk products planned to solve some of these problems but until that's confirmed this picture is pretty accurate. (Note: I am sure there is someone out there running an RA81 in their steambath and has had no troubles, save your typing, a lot of people apparently have had troubles outside a machine room environment and DEC (esp. FS) does not love putting them into anything but machine rooms.) 2. Shared libraries On our 3081 we run a locally built operating system, VPS/VM, which normally only does full loading of libraries at RUN time. You *can* create fully resolved modules (like a.outs) but typically users create a bunch of object decks and a file that looks like: /include loader /include obj1 /include obj2 etc.. and a very fast loader searches and resolves standard libraries (more libraries can be added obviously with more lines) and starts the program. One could imagine the same becoming a norm on UNIX with either a Makefile or a #!/bin/ld variant that took its args on the standard input, locking ld into memory would probably be necessary and possibly looking around to see if some cycles could be saved (and maybe just reserve this to fast systems.) This may not quite be shared libraries but I think it covers most of what the suggestors were after. ld would have to also be ld+go [P.S. I am not claiming VPS to be the only system with this, I think other IBM o/s's had this]. Better would be an ld that could do it directly into memory and just jump to it, I believe TOPS-20 has good tools for this kind of approach where you can allocate a fork that doesn't start and then the parent can set up it's memory image and then start it. Oh well, again, just a thought. Sorta seems simpler than some of the other suggestions (if you don't want to load it till run time why don't you just not load it till run time...:-) I know, speed freaks will be aghast, but it's really not bad in our experience. And, in the UNIX tradition, seems very easy and non-invasive to implement experiments. -Barry Shein, Boston University