Multi-Core, Multi-OS Scaling Performance

Starting with measuring the time it took for GCC 4.5.1 to build PHP on each of the four operating systems, Fedora 14 was the strong leader. When only using one and two of the Gulftown cores, all four operating systems ran nearly identical, but when using 3/4/6/6+HT cores, Fedora 14 pulled a dramatic finish. CentOS 5.5, PC-BSD, and OpenIndiana all scaled close to the same. The PC-BSD 8.1 performance had dipped when three and four cores were used, compared to CentOS and OpenIndiana, which is the beginning of a trend better illustrated in some of the later test results. Without trouble though, Fedora 14 scaled the best when compiling PHP.

The build-apache test profile had not worked with PC-BSD and OpenIndiana, but between CentOS 5.5 and Fedora 14, the Linux 2.6.35-using OS scaled much better. While not the focus of testing in this article, when Intel Hyper Threading was enabled on the Gulftown processor, the benefits of doubling the number of available CPU threads was not linear or nearly as great when scaling from two cores to four cores or three cores to six cores.

With the multi-threaded C-Ray ray-tracing benchmark, the performance between CentOS, Fedora, PC-BSD, and OpenIndiana were identical when the results were normalized against their single-core performance. However, the results did differ when it came to flipping on Hyper Threading. The Core i7 970 with Hyper Threading did dramatically better under OpenIndiana (it was near linear) than it did when Hyper Threading was flipped on in CentOS, Fedora, and PC-BSD where it was of limited benefit.