Cache Compression x MiBench
I just did some tests using newest Compressed Cache, and MiBench (paper here). These tests are part of my Master’s work and were done to evaluate if MiBench benchmarks are suitable for Compressed Cache usage. Since MiBench is largely used in Embedded Systems studies, is important to see if they have memory consumption and/or if the bencharmks are ok to test swap consumption.
MiBench benchmarks are divided in six modules: Automotive, Consumer Devices, Network, Office Automation, Security and Telecommunication. Each module groups programs those act as benchmarks to evaluate common characteristics. What I did was pick one program from each module and execute it in a N810 with Maemo Diablo system. See figures below:
Automotive benchmark
Consumer Devices benchmark
Network benchmark
Office Automation benchmark
Security benchmark
Telecommunication benchmark
As we can notice, graphs presented itself quite similar. It happened due to the memory consumption behaviour applied by each benchmark. Conclusion: MiBench does not do usage of swap area and/or memory. It indicates that we need a better benchmark if we want to evaluate Compressed Cache.
Actually, this is just a small part of my Master’s. Tests are more “real-world” oriented, involving applications like browser, media player and PDF reader.