Lusca Benchmarks
Lusca specific benchmarks will appear shortly. The cacheboy benchmarks should provide a basic idea of how well the code will perform in production. Note that these workloads are for forward proxies serving mainly small files.
Lusca Benchmarks - Cacheboy
How does Cacheboy run in the real world?
This is a work in progress! Benchmark results and real-world performance information will be posted here.
Polymix-4
These profiles are generated with Squid/Cacheboy running on a Dell Poweredge 2650 w/ 2 x dual core 3.06GHz Xeon CPUs, 4GB of RAM, 5 x 36GB U160 10,000 RPM Sun SCSI disks configured as individual drives w/ 4 cache drives and one log/OS drive.
TODO:
- Setup more accurate disk IO, network IO, CPU, RAM monitoring; report that with polygraph results
- Repeat tests with COSS, with more disks, with more polygraph hosts to generate > 1000 req/sec load.
Results
- 27-July-2008: CACHEBOY_1.4.PRE3 Test 1 - first successful polygraph run. ~ 500 req/sec; ~ 30mbit. Poor hit rate is due to local test patches which I forgot to back out!
- 29-July-2008: CACHEBOY_1.4.PRE3 Test 2 - ~ 500 req/sec; ~ 30mbit. Fixed the regression I introduced - hit rate is back to normal! The spikes seen each hour are the hourly log rotation. I'm not sure why this fill phase took so much longer than the previous fill phase. CPU graph available here.
- 7-December-2008: CACHEBOY_1.5.2 Test 1 - ~ 400 req/sec; ~20mbit. It was meant to be 500 req/sec. The testing servers are Linux/FreeBSD instead of both FreeBSD so the workload results may slightly differ to the above.
- 9-December-2008: CACHEBOY_1.5.2 Test 2 - ~ 500 req/sec; ~30mbit. The testing servers are Linux/FreeBSD instead of both FreeBSD so the workload results may slightly differ to the above.
