PROGRESS REPORT (6/8)

■ THIS WEEK - WHAT I DID (6/2~6/8)

SigMP seminar (6/4)

Self-Optimizing Memory Controllers: A Reinforcement Learning Approach (ISCA 2008)

Virtual Hierarchies to Support Server Consolidation (ISCA 2007)

Phoenix

I wanted to test performance using 4 cores (vc105 machine), but there was a problem.

To change a cpu option, the machine has to be rebooted with maxcpus boot options.

I asked the manager of the server to reboot with the options, because I couldn't do that remotely.

The manager said, he is still finding a way to do that.

instead of waiting,

I tested Phoenix version 1.0 on Sun T2 machine (cst5210 machine) with 8 cores (64 threads)

the result graph is in the file attached.

for every applications, the results using mapreduce are no better than p-threads.

compare to CMP speedup in the paper, using 8 cores (32 threads), 

even reverse_index and word_count didn't perform well using mapreduce.

I think when we have many workers, the overhead of mapreduce increases.

the approach in the paper was fresh, but I'm not sure it would be good for the future multicore/multiprocessor systems.

Simics & Gems

since the nodes from Prof. Moon do not have enough space, I got an account of pcal6 machine from a student of Prof. Huh, 

now I am sharing pcal6 machine with JeongSeob.

I downloaded and installed Simics 3.0 on pcal6 again.

SigSimics meeting was on Thursday.

I applied for studying PARSEC benchmark together.

I am trying to figure out the characteristic, and understand the difference between PARSEC and SPLASH2.

http://pcal0.kaist.ac.kr/pmwiki.php?n=WeeklyNote.Infrastructure

Paper

The SPLASH-2 Programs: Characterizationand Methodological Considerations

The PARSEC Benchmark Suite: Characterization and Architectural Implications

PARSEC vs. SPLASH-2: A Quantitative Comparison of Two Multithreaded Benchmark Suites on Chip-Multiprocessors

■ NEXT WEEK - WHAT TO DO (6/9~6/15)

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. Setup Gems simulator.

run SPLASH2 and PARSEC.