On Thu, Jun 11, 2009 at 9:26 AM, Xose Vazquez Perez<xose.vazquez@xxxxxxxxx> wrote: > hi, > > In Apache Benchmark: > "Ubuntu was able to sustain more than 58% more requests per second than Fedora 11" > > <http://www.phoronix.com/scan.php?page=article&item=fedora11_ubuntu904_perf&num=2> Why did you cherry pick the bad news instead of the summary? "In a number of the benchmarks the results were close, but in a few areas there are some major performance differences. In particular, with the test profiles that stress the system disk, Fedora 11 generally did much better -- in part due to the EXT4 file-system and newer Linux kernel. Fedora also did much better with the database tests like SQLite and PostgreSQL. Ubuntu 9.04 though had done a better job with the Apache Benchmark and C-Ray." The disk i/o related tests on the later pages of the multiple page article look really good. And its sort of expected with the move to ext4 in F11 as a default filesystem. We did expect a disk i/o performance for ext4 right? It would be easy to confirm ext4 as a leading factor that by rerunning the same disk i/o and database tests on an Ubuntu system configured to use ext4. The question becomes what is significantly different in the Apache test? Is the Apache test essentially a network i/o test? What is significantly different here that would not end up being tracked to an upstream kernel networking stack regression? Is the apache performance collatoral damage from selinux related latency? Something else in userspace slowing Apache down. I've no idea. I can't imagine its compiler related options on the Apache binaries. -jef -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
