PROGRESS REPORT (8/26)

■ THIS WEEK  (8/17~8/26)

□ Booksim

* single router

single router has 5 inports and 5 outports, so it only has 0~4 source and 0~4 destination.

# of vcs does not affect the result when using no VOQ.

and when # of vcs is 5, the result is very same with the result of 64vcs.

I think because the single router has only 5 destinations (not 64 dests), 

the result will be the same when # of vcs >= 5.

as you said, # of vcs for VOQ should be equal to the # of ports.

* mesh88 uniform

i changed trafficmanagner to use VOQ for injection. so there is no blocking any more.

overall throughput slightly increases, but the graph pattern remains the same.

□ Phoenix

pthread speedup is almost same with the # of processors when network size is < 24.

but the degradation rate is all different using 32, 64 cpus.

for proc = 1 and for some applications, mapreduce takes significantly more time than pthread does.

i am verifying whether the mapreduce code works correctly.

=============================================================

■ NEXT WEEK 

□ Phoenix

test all application and verify whether the mapreduce code works correctly.

□ Booksim

determine why VOQ (single) does poorly on multiflit network.