Intel Linux 3.3 To Linux 3.13 Kernel Benchmarks
The latest kernel benchmarking that happened at Phoronix was testing every major Linux kernel release from Linux 3.3 through the latest stable Linux 3.13 release from an Intel Sandy Bridge system to see how the kernel performance has evolved during the hardware's lifetime for key subsystems.
All of this testing happened from an HP EliteBook that served as Intel's Software Development Platform for Sandy Bridge back in the day. This laptop boasts an Intel Core i5 2520M processor with HD Graphics 3000, Intel 160GB SSD, and 4GB of RAM. All of the hardware was maintained the same throughout testing.
Ubuntu 13.10 x86_64 was the base operating system during testing with Unity 7.1.2, X.Org Server 1.14.5, xf86-video-intel 2.99.907, Mesa 10.1-devel, GCC 4.8.2, and the file-system was EXT4. All of the major Linux kernel releases were obtained from the Ubuntu mainline kernel archive for having a vanilla kernel build and one that is easily obtainable for those wishing to reproduce or validate these results.
All of these kernels were left with their stock configurations and settings throughout testing. Some important notes:
- The Intel HD Graphics 3000 hardware acceleration only worked from Linux 3.6 and newer.
- With the Linux 3.9 kernel is when this Sandy Bridge system switched from using the ACPI CPUfreq driver to using the Intel P-State driver. The default acpi-cpufreq mode on the kernel was "ondemand" while with the P-State driver it's in the "powersave" mode or it defaults to "performance" with Linux 3.13. Changing of the CPUfreq driver also changes how the system CPU frequency was reported to the Phoronix Test Suite system table in going from showing the base CPU frequency to now reporting the Turbo frequency.
- The Linux 3.6 kernel changed from using the CFQ scheduler by default to now using deadline for this Intel solid-state drive.
- Going from Linux 3.3 to Linux 3.4 the default EXT4 mount options changed from "barrier=1,data=ordered,errors=remount-ro,relatime,rw,user_xattr" to "data=ordered,errors=remount-ro,relatime,rw" as used in the rest of the kernels.
Again, this was stock testing of the Linux kernel to represent the experience most Linux desktop users will see that don't invest significant amounts of time tuning their kernels, etc.
All of this Intel Linux kernel benchmarking was handled via the Phoronix Test Suite open-source benchmarking software.