Intel Quad-core: Clovertown Performance Reviewed
Posted on: 11/11/2006 06:00 AM

SPECjbb2005 (Java Server Benchmark) is SPEC's benchmark for evaluating the performance of server side Java. Like its predecessor, SPECjbb2000, SPECjbb2005 evaluates the performance of server side Java by emulating a three-tier client/server system (with emphasis on the middle tier). The benchmark exercises the implementations of the JVM (Java Virtual Machine), JIT (Just-In-Time) compiler, garbage collection, threads and some aspects of the operating system. It also measures the performance of CPUs, caches, memory hierarchy and the scalability of shared memory processors (SMPs). SPECjbb2005 provides a new enhanced workload, implemented in a more object-oriented manner to reflect how real-world applications are designed and introduces new features such as XML processing and BigDecimal computations to make the benchmark a more realistic reflection of today's applications.
SPECjbb2005 is a widely used, industry standard benchmark and it is the second new test in our line-up today. In a nutshell, each "warehouse" spawns an independant thread which determines the concurrency of the benchmark run. Systems tested have a an expected peak number of warehouses (X) that correspond to the total number of "hardware threads" in the machine (i.e. Eight for Clovertown and Dempsey, four for Woodcrest and Opteron). Scores are output as "Business Operations per Second (BOP/s)", and are based on average throughputs of X, X+1, X+2, X+3, etc, up to and including 2X.

I'd like to thank David Kanter (from Real World Technologies) and Larry Gray (from Intel and SPEC) for their guidance in the black arts of SPECjbb2005. This test is a complex one with a myriad of options and switches, and without their help deciphering it all, I wouldn't have been able to include these results here today.

For these tests I used BEA's 64bit Jrockit JVM (jrockit-R26.4.0-jre1.5.0_06-win-x64). I ran the tests with a single JVM and the following commandline options: -server ñXXaggresive -XXtlasize64k ñXms2040M ñXmx2040M

It is easy to see that as thread concurrency builds, having more cores definitely pays off. Clovertown absolutely dominates this test. It is interesting to note that all platforms start to taper off when the number of warehouses approaches 2X, but the rate at which the Clovertown setup declines is a bit steeper than even Dempsey. As I get more familiar with SPECjbb and its commandline options, I'll have to see if I can find out the reason for the rate of decline, and see if I can "performance tune" it a bit to better represent the test platforms' performance.

Printed from (,13.html)