PROGRESS REPORT (9/1)

■ THIS WEEK  (8/27~9/1)

□ Booksim

* 50% of packet; dest 0, 50%: uniform random

voq throughput never exceeds the no voq result.

i counted the number of time when a vc was full and no flit couln't be sent to the vc.

using voq, many flits need to wait for 0 vc.

when bufsize is infinity, throughput of voq is also low.

because the waiting time at 0 vc is very long. 

□ Phoenix

* degradation of map time

i think that the reason of degradation is because of the disk io time.

for shared memory system, cache miss rate can increases when # of processors increases.

when input size gets bigger, the gradation can become large.

actually, L1 cache miss rate increases after 8 threads.