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

List:       hpux-cxx-dev
Subject:    RE: CXX-DEV: Strange phenomenon with +I binaries
From:       "MAUCCI,CYRILLE (HP-France,ex1)" <cyrille.maucci () hp ! com>
Date:       2004-05-22 7:35:34
Message-ID: CF35E3640542D411B8E600D0B747B166131E843C () camus ! grenoble ! hp ! com
[Download RAW message or body]

Hi Carl,


I think you've found the origin of the problem!
In one case (the long time operation) I was launching my command from a NFS
mount point, in the other case I was not.

Thanks a lot!

Cyrille

-----Original Message-----
From: Carl Burch [mailto:cdb@cup.hp.com] 
Sent: samedi 22 mai 2004 01:11
To: sandyam@india.hp.com; cyrille.maucci@hp.com
Cc: cxx-dev@cxx.cup.hp.com
Subject: RE: CXX-DEV: Strange phenomenon with +I binaries



Cyrille-

> I have actually made further tests with and without this variable set 
> to ON/OFF and I have seen that without exporting ENABLE_PBO_FORK to 
> OFF it can take up to 10 minutes before flow.lock is released... This 
> is really a long time since it is 10 minutes of flow.data generation 
> while executable ran only ... 3 seconds...

   I didn't see anyone mention the point of the fork in the flow.data writer

code - it's to let your original process complete so that logic (e.g., from 
a driver program) ensuring the child process dies on schedule doesn't "kill
-9" the process while the flow.data file is half written.

   While that helps ensure you don't get truncated flow.data files, it
doesn't help with the original problem of it taking a long time to write out
the flow.data file.  That's often the result of writing the flow.data file
to a slow NFS mount, so you might try setting the FLOW_DATA environment
variable to point at a local filesystem like /tmp (where you have enough
disk space, of course).

> I have given a try to tusc and will send you the output (offline).

   If my guess doesn't work, then that's the right answer - send the tusc
output to Sandya. ;-)

							- Carl
							
 _________________________________________________________________
 To leave this mailing list, send mail to majordomo@cxx.cup.hp.com
    with the message UNSUBSCRIBE cxx-dev
 _________________________________________________________________
[prev in list] [next in list] [prev in thread] [next in thread]