Why the SPARC architecture is dead
The project (Java application, consisting of a number of server processes) I work on currently uses a Sun Fire 15K server in production (32G RAM, 12 CPU). Today we decided to test the same application on a Sun Fire v40z server (AMD Opteron 4 CPU, 32G RAM).
Running the same performance test on both server configurations, we found that the Opteron server was 3x faster than the SPARC server. Another group have done a similar server comparison but using there multi-process multi-thread C++ server and found a 3-4x speed improvement.
It's quite interesting to compare the cost of both systems - 15K vs V40z.
Given the performance improvement and cost savings, it's difficult to see where Sun is going with the SPARC architecture.
1 Comments:
Maybe your app is not utilizing all the horsepower of the Sun Fire 15K CPUs.
I mean, you may get the same result by removing some of the CPUs from the motherboard.
How did you test the system? How many simulated concurrent users where using it?
Post a Comment
<< Home