*BSD News Article 6271


Return to BSD News archive

Newsgroups: comp.unix.bsd
Path: sserve!manuel.anu.edu.au!munnari.oz.au!sgiblab!swrinde!elroy.jpl.nasa.gov!usc!venice!gumby.dsd.trw.com!eel.dsd.trw.com!gottloeb
From: gottloeb@eel.dsd.trw.com (Jeffrey R. Gottloeb)
Subject: Re: The Compile from HELL!
Message-ID: <1992Oct9.005309.13915@gumby.dsd.trw.com>
Keywords: 386BSD GCC compiler Xview3
Sender: news@gumby.dsd.trw.com
Organization: TRW, INC.
References: <1992Oct6.074223.615@coplex.com>
Date: Fri, 9 Oct 1992 00:53:09 GMT
Lines: 50

In article <1992Oct6.074223.615@coplex.com> chuck@coplex.com (Chuck Sites) writes:
>
>
>    Recently when trying to port XView3 to 386BSD and XFree, I ran into a
>brickwall of a performance problem.  There are some files which start 
>compiling but they just don't stop.  One file, attr_char.c, I let compile
>over the weekend only to come back and find the system apparently frozen.
>Frozen may be a misnomer though.  While trying to telnet into the system 
>during the early stages of the compile, I can access the system, but slowly
>the perfomace degrades to the point of printing 1 line/min for an ls, and it
>gets worse to the point of not responding at all.  I assumed it was swapping,
>but the severity that other process are excluded from thier share of the
>clock makes me think this is a kernel problem. 
>
>    Well in fairness, this may not be a kernel problem alone. GCC maybe
>partially to blame.  On large structure arrays I've seen it go 'all
>consuming' before on OS2 and SYSV,  but both resulted in an out of
>memory error at the user prompt after several hours. Neither had the
>multitasking performace degradation like I got with 386bsd.  (I was
>trying to port the fax portion of the TIFF package on uunet fyi.) 



I have seen this problem also.  On my system, I have 16MB of swap space
and set limit datasize 14m.  CC aborts with virtual memory exhausted while
processing g3states.h.  This is a hugh file that seems to only have data
initialization code in it.  I have noticed using the patched version of ps
that the vsz attribute of the compile process continues to climb to the 14MB
value causing the abort.  I am not sure if cc is working correctly and
trying to keep the arrays in memory as it is compiled or not.


Jeff Gottloeb
gottloeb@gumby.dsd.trw.com



>
>    So, my question is: has anybody seen this type of problem and is this
>symtom fixed by the latest beta patches?  Second, if the beta patches do
>fix this problem, could someone direct me to a site that has the collected
>binaries of the beta patchs.        
>
>
>Thanks,
>Chuck Sites         
>chuck@coplex.com    (502)-9688495
>