Charm++ parallel programming system

["Kale, Laxmikant V" <kale AT illinois.edu>]

(In addition to Phil’s advice)

How big is the trace data with 19 queens? (are you using grain size control?)

Also, I am not sure why you’d want detail trace data for a divide-and-conquer program. Maybe summary data is enough?

-tracemode summary.

-- 

--- 

Laxmikant (Sanjay) Kale         http://charm.cs.uiuc.edu

Professor, Computer Science     kale AT illinois.edu

201 N. Goodwin Avenue           Ph:  (217) 244-0094

Urbana, IL  61801-2302          

From: Hrushit Parikh <parikhhrushit AT hotmail.com>
Date: Sunday, February 15, 2015 at 7:18 PM
To: "charm AT cs.uiuc.edu" <charm AT cs.uiuc.edu>
Subject: [charm] Unable to run UTS with projections

Hi,
I am trying to get projections analysis on the UTS benchmark (Tree T1L) in charm++ using the "-tracemode projections" option. I use the workstealing seed balancer.

I am using charm-6.5.1 version on the UT Austin Stampede cluster. The stampede cluster uses ibrun to launch the executable. I have compiled charm using ./build charm++ mpi-linux-x86_64   mpicxx  -j16.

With 256 Processors, the T1L run takes only 0.3 seconds when I compile charm with the "--with-production" option (to disable traces). However, with traces enabled the execution time exceeds 15 minutes and I have to kill the program. 

I tried running NQueens with the same workstealing balancer on 256 Processors with traces enabled. For queens upto 19 (grainsize=3) it runs fine and projections data is generated, but with 20 queens it takes forever.

Any ideas why UTS takes forever with traces enabled or am I missing something?

Thank you,
Hrushit