Hi all,
I'm running M5 in Full-System mode with 8 CPUS. I've applied the Linux patches
for BigTsunami support up to 64 processors.
The simulation works well using PARSEC multithreaded benchmark "blackscholes"
with small inputs and 64 threads.
In the next step, I'm trying to run with multiprogram workloads on more
processors and having the following questions:
(1) In M5 FS mode, is there a way to assign different multithreaded
applications to different groups of CPUs?
For example, if I have 8 CPUs, and two multithreaded benchmarks (
"blackscholes" and "canneal") each with 64 threads.
I'd like to assign "blackscholes" to cpu0 to cpu3; and assign
"canneal" to cpu4 to cpu7; to let them run in parallel.
Is there a way I could do that? If there is, how could I configure it
in the simulation script?
I've learned how to config multiprogram workloads in SE mode. But in FS
mode, should I also consider changing the .rcS script?
(2) Also, I'm trying to increase the number of CPUs to 64. (All the
configures are the same expect I'm using "-n 64" in the command line.)
Firstly, I created a checkpoint at the beginning of the ROI (region of
interest--parallel phase), to first run M5 in simple mode.
Then, I run M5 in detailed mode beginning at the checkpoint.
Everything works fine in the first step. But there's an "Segmentation
fault" as follows when restoring the checkpoint.
*****************************************************
Restoring from checkpoint
warn: Not unserializing 'system.cpu54.dcache': no section found in
checkpoint.
For more information see: http://www.m5sim.org/warn/dc3e87f8
warn: Not unserializing 'system.cpu54.icache': no section found in
checkpoint.
For more information see: http://www.m5sim.org/warn/dc3e87f8
Segmentation fault
*****************************************************
I've been using exactly the same configuration and steps with 8 CPUs
and it works well. I was wondering why it doesn't work with 64 cores.
Is there anybody met the same problem before? Or has any idea what may
cause this problem?
Any comment and suggestion would be appreciated,
Thanks in advance,
Jie
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
