home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.prime
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!rpi!rpigate!x
- From: J.BUSH@MD-B.Prime.COM (John Bush, PrimeService +1 508-620-2800 Voice x7068 Fax x5744)
- Subject: Re: tuning
- Message-ID: <9208282357.AA20274@usenet.rpi.edu>
- Sender: ListServer@Blx-A.Prime.COM (Prime Netherlands ListServer v1.12)
- Reply-To: Info-Prime@Blx-A.Prime.COM
- Date: Fri, 28 Aug 1992 23:54:44 GMT
- Lines: 84
-
- RE: 4150 tuning
-
- It's been a while since I was directly involved in this,
- and I need a respite right now ...
-
- > .... There are about 30 users logged in, all using the same
- > login id and therefore the same CPL that presents a menu.
-
- Oh no, not another LOGIN.CPL menu system! I'm not sure when
- a fix went in for this... people from Product Integrity can
- clarify it. I'm working off memory here. Up until a recent
- rev of Primos, the timeslice granted at Login was real BIG.
- The rationale was to be able to complete external login processing
- before hitting the low priority queues. By kicking off the APP
- from within LOGIN.CPL, the system thinks you're still logging in,
- and your timeslice is something like 32000 seconds!
-
- Thus, break loose from it. If you have an external login program,
- try inserting a CALL RECYCL in it. The RECYCL call actually
- throws yourself at the mercy of the scheduler. But as a result,
- you come out of it with a new (regular) timeslice. It may take
- 5 extra seconds to login if the system is busy, because of the
- queuing on the low pri queue. This trick is untested as far as
- I know. Plus, it's listed as "OBSOLETE" in the Subr. Ref. Guide.
- Ha! It's been obsolete for several years now, but keeps right on
- ticking. I used to use RECYCLE to time scheduler performance
- and calculate queue service rates.
-
- Another thing you can try is to force the LOGIN.CPL to exit without
- losing menu control. Here's how: Extract the menu stuff out of
- LOGIN.CPL and put it into a program called USERMENU.CPL.
-
- Create a "BREAKAWAY.COMI" file that looks like the following:
-
- R <wherever>USERMENU
- CO -TTY
-
- Then, terminate LOGIN.CPL like this:
-
- COMI <wherever>BREAKAWAY.COMI
- &RETURN
-
- This convoluted technique uses command processor "abeyance" to
- force execution of the "&RETURN" statement in the LOGIN.CPL
- program. When you issue the COMI command, it doesn't force
- execution of the file right away, it merely redirects the command
- stream so the next user command will come from that source. CPL
- doesn't read directives from the command stream. However, as soon
- as the &RETURN gets executed, the next command will be picked up
- from BREAKAWAY.COMI, which in turn, gets you into the menu, but
- you are completely free from the original LOGIN.CPL.
-
- > .... The problem seems
- > to be when the user has to stop one .RUN program and return to the CPL
- > menu to invoke another .RUN.
-
- Try creating REGISTERED EPFs.
-
- In addition, modify the user profile: Give more active EPFs per user,
- and more dynamic segments. Do be careful, however, because your system
- is paging, and anything which adds to the memory demand can be
- detrimental. (is that the same as "decrimental"?)
-
- Investigate what is causing the your page faults.
-
- Maybe you have seek contention, with PAGING and program data residing
- on the same drive. Can you relocate PAGING to another less used drive?
- Do you have too many LOCATE BUFFERS configured? I used to advocate
- always setting locate buffers to the maximum, simply because if an
- I/O was to occur, it doesn't matter whether it happens as a page fault
- or as a file system read. It might actually be more efficent to
- do a page fault in response to a LOCATE FIND rather than to MISS and
- do regular I/O. On the other hand, LOCATE buffers can thrash against
- large programs rolling in and out of memory. Trial and error is still
- the best method for figuring it out, as long as you pretend you know
- what you're doing. Depending on the reason for the page faults, adjust
- accordingly. (or do I mean, "adjust the accordion?")
-
- If page faults were due to your programs' working set size, adding
- memory should have helped.
-
- Regards,
- John Bush
-
-