Newsgroups: comp.parallel.pvm
From: patton@phy.uab.edu (David Clay Patton)
Subject: Re: How many tasks can one spawn on a single host
Organization: University of Alabama at Birmingham
Date: 17 Nov 1995 16:46:17 GMT
Message-ID: <48ie8p$n81@maze.dpo.uab.edu>

Tom Cox (coxt@tempest.adsnet.net) wrote:
: Chetan N Vora <chet@brc.uconn.edu> wrote:

: >Hi everyone,

: >I am trying to build a parallel MD code using the master-slave paradigm.
: >Currently, I am using two Sparcstation 2000 with 8 cpus each connected with
: >Ethernet. I am having the following problems.

: >Problem 1: My code runs fine when I run it on a single host and spawn less than
: >6 slaves processes.But it gives me the libpvm error -
: >"libpvm [task number]: peer_conn() shmat to t40006: Too many open files" 
: >when I spawn more than 5 tasks.

: >What does this message mean ?? I get this error only when the no. of slaves
: >exceed 5 ; the program does go to completion but doesn't give correct results
: >(because of incomplete computations, I guess.) Is there any heuristic metric
: >about how many tasks should one spawn on a host ??

: >Problem 2: When I add the other host to the configuration and try to spawn
: >tasks using PvmTaskDefault, the code just hangs. 

: >Symptoms:
: >The log file on host on which I started the console shows the tasks to be
: >spawned with some task(s) having the error
: >"libpvm[task number]: mroute() frag with no message".

: >It appears that this task(s) doesn't receive at all the stuff that the master
: >sends it.

: >Is there any documentation about the meaning of the error messages that are
: >returned in the log file ??

: >Any help will be great appreciated.

: >Thanks in advance,
: >Chet

: > 
: >-- 
: >*******************************************************************************
: >Chetan Vora		www: http://www.eng2.uconn.edu/~chet
: >The scientific theory I like the most is that Saturn's rings are composed 
: >entirely of lost airline baggage.  -Mark Russel
: >*******************************************************************************

: We currently have the same problem on our 4 processor sun here at CWU.
: We get the exact same reported errors if we try to spawn 5 or more
: processes.   Any ideas?

: -tom cox
: coxt@tahoma.cwu.edu
: Systems Analyst/Programmer I


In $(PVM_ROOT)/Readme.mp it says:

Solaris2.4
----------

APPLICATION PROGRAMS

        All PVM programs must be linked with the thread library (-lthread).
Refer to the example Makefile (pvm3/examples/SUNMP/Makefile) for
details.

BUGS AND CAVEATS

        a) There is a system limit on the number of shared-memory segments
a process can attach to. This in turn imposes a limit on the number of
PVM tasks allowed on a single host. According to Sun Microsystem, the
system parameter can be set by a system administrator:

=====================================================================
Here is the information on getting the limits:
 
One way if this is what you want.
Must be root.
gwhiz# adb -k /dev/ksyms /dev/mem
shminfo,4/D
shminfo:
shminfo:        268435456       200             200             200
 
 
So shmseg is the fourth field which is 200 decimal.
Remember that this driver is only loaded if you use it. So shminfo will
show up symbol undefined unless the driver has been loaded.
 
The default seems to be six but for tpc we up it in "/etc/system".


--
---------------------------------------------------------------------------
David Clay Patton                                  patton@phy.uab.edu
Physics Department                                 FAX:205-934-8042
University of Alabama at Birmingham                
Birmingham, Al 35294                     http://www.phy.uab.edu/~patton
---------------------------------------------------------------------------
"It is a miracle that curiosity survives formal education." -Albert Einstein

