[prev in list] [next in list] [prev in thread] [next in thread] 

List:       hpux-cxx-dev
Subject:    RE: CXX-DEV: Memory consumption (was Re: aCC 1.09: Assertion failed...)
From:       Vijay Rajagopalan <vijay () roguewave ! com>
Date:       1998-02-23 17:58:41
[Download RAW message or body]

{}To be blunt, aCC is a memory hog.  The process sizes of ctcom seemed to
{}grow to about 55-65MB for every file I complied -- and this is without
{}optimisation (I dare not even think how much more memory it will want if
{}I turn optimisations on).  Given that I have only 64MB of real memory,

In our case, aCC 1.09 exited with "Out-of-memory"  (with -g option), 
exit code 2 or exit code 3 (with -O option) when compiling certain
files.  We had to increase the "datasize" parameter of the limit
command (in csh) from 65 MB to 132 MB. (The machine we
used also had only 65 MB of main memory.)

Please note that the same affected files compiled well with 
aCC 1.06 under the same old environment (with datasize of 65MB!).

//Vijay

-----Original Message-----
From:	Lassi A. Tuura [SMTP:lat@iki.fi]
Sent:	Monday, February 23, 1998 3:33 AM
To:	cxx-dev@cxx.cup.hp.com
Subject:	CXX-DEV: Memory consumption (was Re: aCC 1.09: Assertion failed...)

Richard J. Rauenzahn wrote:
|> This also error'ed on my machine -- aCC 1.09, HPUX 10.20
|> 
|> I found the assertion went away if I didn't use the precompiled header
|> file and just added #include "fast-stdlib.cxx" to the top of
|> ATLASModelCreator.i..

Thanks, this indeed did solve the problem and I managed to finish the
build.  Looking at the dog slow build to progress, I started to pay
attention to `top' output.  First, doing nothing else on my machine but
running make (no parallelism), the machine runs at load levels over
three. Of that, it's about 80% idle time, `vhand' being the most active
process :-( Trying to do anything else on the machine while it builds is
futile; just bring one xterm window on top of another one takes about 10
seconds. 

To be blunt, aCC is a memory hog.  The process sizes of ctcom seemed to
grow to about 55-65MB for every file I complied -- and this is without
optimisation (I dare not even think how much more memory it will want if
I turn optimisations on).  Given that I have only 64MB of real memory,
and that aCC apparently wanted to actually touch much of what it
allocated, the machine ended up trashing pretty badly.  It seemed that
the build times were faster with precompiled headers, but that may just
have been an impression; because of the above bug I could not test them
in this part of the application and I did not look at the stats in the
other parts that carefully. 

So, are the any improvements to be expected in this area, or should I
have more memory installed on this machine (and any others we use with
aCC)?  Are there any kernel parameters that could be tuned? In the past,
I have tried to make the dynamic file system cache as small as possible
and it has significantly reduced trashing.  I am not sure I can shrink
the cache any more.

In case it matters, this is HP-UX 10.20, C100 (aka 9000/777).

//lat
--
Every old idea will be proposed again with a different name and a
different presentation, regardless of whether it works. --RFC1925

 _________________________________________________________________
 To leave this mailing list, send mail to majordomo@cxx.cup.hp.com
    with the message UNSUBSCRIBE cxx-dev
 To reply to the list, send mail to cxx-dev@cxx.cup.hp.com
 _________________________________________________________________

 _________________________________________________________________
 To leave this mailing list, send mail to majordomo@cxx.cup.hp.com
    with the message UNSUBSCRIBE cxx-dev
 To reply to the list, send mail to cxx-dev@cxx.cup.hp.com
 _________________________________________________________________

[prev in list] [next in list] [prev in thread] [next in thread]