by Oleg Dulin
One of the most effective ways to measure raw CPU speed is to run a real-life computationally-intensive program whose memory footprint is small enough to fit into the CPU cache. The program must run at a low priority during 24-48 hours to account for interference common to multiuser environments such as Linux, which runs on NorthCountry. RC5-64/DES decryption program is an example of such algorithm.
To perform the test, I had installed RC5 Personal Proxy onto the controller node. Since the proxy is not CPU intensive and simply provides a means for nodes to communicate with the central key server, I decided to not only run RC5 clients on all four nodes but also on the controller machine. With this setup, all six of the CPUs were 100% loaded. The experiment ran for approximately 50 hours, 16 of which were during the day.
According to RC5 proxy reports, NorthCountry processed 3.85 Millions of keys per second (MKeys/sec). Using distributed.net statistics reports and a simple proportion it is easy to calculate that NorthCountry operates at 357.39 MFlops-3. As described on Performance Database, MFlops-3 weightings are: FADD 42.9%, FSUB 3.4%, FMUL 50.7%, FDIV 3.4%.
It is interesting to note that NorthCountry performance was between Digital Alphasystem 4000 and FujitsuVP2200/10 multiprocessors. That is between top 8th and 9th place on Performance Database chart. With only six 233 Mhz Pentium II CPUs and a total of 512 MB of RAM, NorthCountry is not the fastest supercomputer around. However, considering the total cost of the machine its price/performance ratio cannot be beaten.
Since NorthCountry is highly scalable, adding only four 233 Mhz Pentium II nodes will increase the performance to beyond 1 GFlops-3.
by Boris Jeremic
Look here for NAS Parallel Benchmark results (class W and S) for four nodes of NorthCountry.