PROGRESS REPORT (6/15)

■ THIS WEEK  (6/9~6/15)

□ Phoenix 

Phoenix_version1.0.pdf

(1) Unit size

I checked the unit_size of each application.

They tested unit_size with 4KB, 16KB, 64KB, and 128KB in the paper, but the default size of each application is as below.

they calculated chunk size using the unit size and L1 cache size. (chunk_size = unit_size / L1_cache_size)

but their code cannot get L1 cache size from system environment, and always use default size as 64KB.

I will get the real cache size and make a graph again. 

(2) # of threads

I am not sure whether the threads are divided equally among processors or not. 

I tried to find out the way, but couldn't find the system command on Solaris.

(3) Sequential time

Sequential time results of each application are almost same, but they are not perfectly equal.

i think it's because the machine state is a little different each time.

(4) One processor speedup 

the result graph is attached at the file. 

the processing times of mapreduce and pthreads are similar to the sequential, 

but mapreduce performed better for some implementation (ex. wordcount, reverseindex). 

□ SigMP seminar (6/11)

Coordinated Management of Multiple Interacting Resources in Chip Multiprocessors: A Machine Learning Approach (Micro 2008) 

Bandwidth adaptive snooping (HPCA 2002)

□ Simics & Gems

Study and get used to Simics. run SPLASH2 and PARSEC.

■ NEXT WEEK  (6/16~6/22)

□ SigMP seminar (6/18)

An OS-Based Alternative to Full Hardware Coherence on Tiled CMPs (HPCA 2008) (by 김대훈)

PARSEC vs. SPLASH-2: A Quantitative Comparison of Two Multithreaded Benchmark Suites (by 신민정) prepare Seminar

□ Simics & Gems

Setup Gems simulator. run SPLASH2 and PARSEC.