The First Benchmarks Of Red Hat Enterprise Linux 6.0
Beginning with the Apache benchmark we actually see a huge performance loss between CentOS/RHEL 5.4 and Fedora 12 / RHEL 6.0. The performance of Ubuntu 10.04 LTS is also a ways behind CentOS/RHEL 5.4. The difference that these newer file-systems carry is the EXT4 file-system by default rather than EXT3. Among the updated distributions, however, Canonical's Ubuntu 10.04 did the best with its Linux 2.6.32 kernel. As we have now covered in many EXT4 articles, there are a number of performance regressions to be found with EXT4 in recent kernel releases as these updates bring better file-system integrity. It is only now though that the enterprise-grade distributions are switching to these newer kernels and moving to EXT4 by default.
The PostgreSQL performance also takes a hit in the newer distributions, with the exception of Fedora 12 that is running shockingly faster. In fact, Fedora 12 ended up being 4.5x faster than Red Hat Enterprise Linux 6.0. Most of this speed-up though is likely attributed to Fedora 12 shipping with the older Linux 2.6.31 kernel by default, where as the Linux 2.6.32 kernel carries an ext4_sync_file change that significantly hampers the file-systems performance, as we autonomously found the regression in the Linux kernel. Again, these slowdowns in the other distributions can be attributed to EXT4. Between Ubuntu 10.04 LTS and Red Hat Enterprise Linux 6.0, the two distributions were performing the same in this database test.
With the Java-powered Bork file encryption benchmark, the new Linux distributions bring measurable performance gains. It took 95 seconds to encrypt the sample file in CentOS, but it dropped to 59 seconds in the Red Hat Enterprise Linux 6.0 Beta. Meanwhile, Ubuntu 10.04 LTS took just 47 seconds to complete this operation.
Compared to CentOS 5.4, RHEL 6.0 Beta regressions in the C-Ray performance. However, Red Hat Enterprise Linux 6.0 Beta is not any slower than Fedora 12 or Ubuntu 10.04 LTS with this multi-threaded ray-tracing benchmarking.