Charm++ parallel programming system

[Ryan Mokos <ryan.mokos AT gmail.com>]

Hi Xuehan,

This is odd.  While such a difference between the OVTs and GVT could theoretically exist, it should only last until the end of that GVT iteration.  It is also possible for the OVTs of some of the posers to be less than the GVT if they haven't had recent events processed on them.  However, I think that the OVT of at least 1 poser should always be >= the GVT.  In any case, your OVTs look reasonable while the GVT does not.

First, are you using 64-bit timestamps in POSE?  To check if you are, look in charm/src/libs/ck-libs/pose/pose_config.h and make sure the line with #define USE_LONG_TIMESTAMPS 1 is *not* commented out.  You want USE_LONG_TIMESTAMPS set to 1.

Next, are you printing the GVTs and OVTs correctly?  Are you treating them as 64-bit integers?

Finally, do you treat timestamps in your network model as 64-bit integers?

If looking at those things doesn't point out any problems, try running the simulation sequentially (compile BigNetSim with SEQUENTIAL=1 and run on 1 processor) and see if you get the same results.  You could also try looking at conGVT (conservative GVT), optGVT (optimistic GVT), and estGVT (GVT estimate) in charm/src/libs/ck-libs/pose/gvt.C and print their values to see if you can find at what point the GVT blows up.

Ryan

On Mon, Oct 31, 2011 at 8:33 AM, Xuehan Xu <xxhdx1985126 AT gmail.com> wrote:

Dear Sirs:
       I'm trying to build some network model on top of POSE, but I got an overwhelming problem: my simulation time in terms of GVT ticks is very large-----such as: 595541780309723496. To find out what went wrong, I printed out all the OVTs of all the posers. But all these OVTs seem to be normal-----like 283340332.

       Why can the GVT be so large, while the OVTs are all small?

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

_______________________________________________
ppl mailing list
ppl AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/ppl