Mike,
Thanks for the information. I work with Jim and am also trying to
understand the differences in memory use of the z/VM 5.3 TCP/IP
clients compared to the z/VM 4.4 ones.
The web page you reference talks about C sockets library changes, but
says these were already present in z/VM 4.4. The difference we're
seeing is between 4.4 and 5.3. So does this apply? I also wonder
because we're seeing this with ftp and telnet, which aren't mentioned
on your web page as using the C socket library. (Do they?)
I've done some more testing, and find that the 5MB or so additional
memory use that we're seeing only happens when there is a DNS lookup
involved. So if I do "ftp cornellc.cit.cornell.edu" I see the
additional memory use, but if I do "ftp 132.236.98.12" I
don't. Similarly "ftp loopback" doesn't cause the additional memory
use. I see the same thing for lpr: lpr profile exec (p raw at
cornellc.cit.cornell.edu causes the additional memory use, while lpr
profile exec (p raw at 132.236.98.12 does not.
So what is there about DNS lookup that causes this memory use, while
otherwise the clients don't suck up virtual storage?
Thanks,
Mark Bodenstein ([EMAIL PROTECTED])
Cornell University
At 01:08 PM 9/14/2007, you wrote:
Jim,
There are several reasons for the increased use of virtual storage
when running
the TCP/IP functions and utilities. You can read about some of these at
<http://www.vm.ibm.com/devpages/donovanm/zvmle.html>http://www.vm.ibm.com/devpages/donovanm/zvmle.html
and specifically look at
<http://www.vm.ibm.com/devpages/donovanm/zvmle.html#LEstor>http://www.vm.ibm.com/devpages/donovanm/zvmle.html#LEstor
The 5M "feechur" you mention is an unfortunate artifact of sockets
being defined
as POSIX file descriptors in the Byte File System client code. "Fixing" this
involves a significant rewrite of a BFS client storage management and will
most likely not happen any time soon.
Mike Donovan
---
zVM 5.3 TCPIP memory problem
I remember back in zVM 5.1 or 5.2 days seeing on the list that there
were memory problems or memory size issues. I probably didn't follow it
because I was using 4.4 and probably just thought it would be fixed.
Now I see that it wasn't fixed. I've just been told by IBM it's a
"feechur". I'm seeing it with the TCPIP clients, FTP and LPR. They
grab about 5M and don't release it. I didn't see it in putting the
release together because whoever runs the MAINT id in installation with
a small machine size.
Does anyone have any ideas or a solution or is that just the way it is?
Jim
--
Jim Bohnsack
Cornell University
(607) 255-1760
[EMAIL PROTECTED]