RALPH@UHHEPG.BITNET (03/16/88)
Date: 15-MAR-1988 22:16:16.19 From: Ralph Becker-Szendy RALPH AT UHHEPG To: 0::"info-vax@kl.sri.com",RALPH Subj: RE: System memory tuning, and please help with a modem program Hi everyone. Sorry if you see it for a second time. It seems it never made it out (i never saw it, and never got any replies) ... Here the very short summary of answers to my question of a while ago: we have a 11/780 with 16MB, typically 15 users and some batch jobs, but half the memory seems to be used, and there is heavy paging, which makes the machine real slow. The main part of the answer to the problem is simply to make the workspace for users and batch jobs a lot bigger, the 1000 pages maximum be had before was just too restrictive. We are now running with WSEXTENT=2000, and are down to 10 soft page faults, and less then 1 disk page fault per second on the average; but we are going to go higher in steps (to see when it "feels" best). It also matters where the pagefile is: it makes sense to make the RP07 on the massbus (rather than the RA81 on the unibus) the paging disk, but we haven't implemented that yet (disk space too tight, we can't just shuffle 20000 blocks like that). Concerning SYSGEN parameters, a lot of people suggested increasing MPWHILIMIT. We will try this later. As you may see, we are tuning the system in small steps, so as not to upset users or the system. Thanks very much for all the help (some of which even came by US mail, which tells you about the reliability of INFO-VAX). Now two other little questions: First, we are going to upgrade from 4.2 to 4.6 very soon (we should have done that long ago, just never got around to it). In terms of big packages we are running Mass-11, VAXsim-Monitor, SMP, ANL-NJE for the RSCS link, Fusion for the TCP/IP link, and the Talaris font manager (inside the print symbiont). Does anyone out there of any particular problems we might run into with 4.6 or its interaction with these packages ? Typically for several weeks after a VMS version change our system manager has to go around and get all possible things back to work. basically the question is, are there some things we should better be aware of, concerning version 4.6 ??? The second question is more subtle. You all know and love SET/HOST/DTE. It "connects" you to another terminal line, which we use for outgoing calls on modems (our way to get to the mainland, you should see our phone bills). What is missing in SET HOST/DTE is the capability to keep a session log-file part of the time, typically for downloading: dial out via modem, log into the remote machine, switch the log on, type a file out, switch the log off again. Therefore, we use a program called HOST (comes from Fermilab), which does that quite nicely, and is also otherwise easy to use (it does all the allocate, sets the speed, and can even do uploading, but we hardly ever user that). Now the problem is the following: this HOST program is incredibly inefficient, because it issues loads of QIO requests: for a typical session at 2400 baud it is 300 buffered IOs per second (which makes sense speed-wise). As far as i understand the code it actually issues two QIO requests, but instead of just having an AST routine wait for a completion it will keep a timer, and cancel all the outstanding requests every time one request is answered or when the timer runs out ! As a result the CPU becomes just incredibly inefficient: a glimpse into MONITOR MODES shows 5 to 10% of the CPU time is spent in user mode, the rest in kernel mode and on the interrupt stack. Net result: the VAX becomes a virtual machine --- it is virtually dead (hardly surprising, when running at 5% efficiency). We are trying to talk users who don't need session log files into using SET/HOST/DTE instead of the HOST program, but i don't know whether that is a lot more efficient. I have looked into KERMIT, but the frog doesn't have switchable log files either, and it is rumoured that he is very inefficient, too. So the question is: Is there a good "modem"-type program out there, which has switchable session log files, and does not cripple the CPU ? Thanks a lot in advance for your help, and as usual i'll summarize responses to INFO-VAX. Ralph Becker-Szendy RALPH@UHHEPG.BITNET University of Hawaii / High Energy Physics Group (808)948-7391 Watanabe Hall #203, 2505 Correa Road, Honolulu, HI 96822 "Hawaii - it's not just for tourists. People actually live and work there."
carl@CITHEX.CALTECH.EDU (Carl J Lydick) (03/19/88)
> What is missing in SET HOST/DTE is the capability to keep a session > log-file part of the time, typically for downloading: dial out via > modem, log into the remote machine, switch the log on, type a file out, > switch the log off again. Therefore, we use a program called HOST (comes > from Fermilab), which does that quite nicely, and is also otherwise easy > to use (it does all the allocate, sets the speed, and can even do > uploading, but we hardly ever user that). > > Now the problem is the following: this HOST program is incredibly > inefficient, because it issues loads of QIO requests: for a typical > session at 2400 baud it is 300 buffered IOs per second (which makes > sense speed-wise). As far as i understand the code it actually issues > two QIO requests, but instead of just having an AST routine wait for a > completion it will keep a timer, and cancel all the outstanding requests > every time one request is answered or when the timer runs out ! As a > result the CPU becomes just incredibly inefficient: a glimpse into > MONITOR MODES shows 5 to 10% of the CPU time is spent in user mode, the > rest in kernel mode and on the interrupt stack. Net result: the VAX > becomes a virtual machine --- it is virtually dead (hardly surprising, > when running at 5% efficiency). We are trying to talk users who don't > need session log files into using SET/HOST/DTE instead of the HOST > program, but i don't know whether that is a lot more efficient. I have > looked into KERMIT, but the frog doesn't have switchable log files > either, and it is rumoured that he is very inefficient, too. > > So the question is: Is there a good "modem"-type program out there, > which has switchable session log files, and does not cripple the CPU ? How about SET HOST/DTE? It DOES allow what you want. However, you have to allocate the terminal before you issue the SET HOST/DTE command (this is generally a good idea anyway, since otherwise, a buffer overflow kicks you out of SET HOST/DTE, and if it's a BAD buffer overflow, tends to start a process running LOGINOUT on the DTE you SET HOST/DTE to). In general terms, a typical session with selective logging looks like: $ ALLOCATE Tcnn MY_DTE !allocate terminal line $ SET HOST/DTE MY_DTE !SET HOST/DTE without logging Type what you want without logging; end with CONTROL-\ $ SET HOST/DTE/LOG MY_DTE !SET HOST/DTE with logging Do whatever you want logged; end with CONTROL-\ $ SET HOST/DTE MY_DTE !SET HOST/DTE without logging Do what you don't want logged (including, for example, logging out); end with CONTROL-\ $ DEALLOCATE MY_DTE !release terminal line