CentOS vs. Oracle vs. Scientific Linux 6.1 Performance

[phoronix]

Phoronix: CentOS vs. Oracle vs. Scientific Linux 6.1 Performance

While CentOS, Scientific Linux, and Oracle Linux Server are all derived from the same upstream source (Red Hat Enterprise Linux), how does the system performance compare between these RHEL derivatives? Here are some benchmarks of each of the 6.1 releases for Oracle Server, CentOS, and Scientific Linux, as they all do not perform the same.

http://www.phoronix.com/vr.php?view=16835

[Milan Kerslager]

The difference

I don't understand the difference. The kernel from Oracle could be different. But where is the other difference? Maybe the kernel has been compiled by another compiler (not that one in the distro), which should be noted in the text (eg.: cat /proc/version). Maybe tests were affected by some application on background (different set of components has been installed). Without closer answer, the tests are worth nothing IMHO.

[whitecat]

I'm most surprised by the difference between SL and CentOS.

Oracle probably applied some patches (so it's not completely equal to RHEL) in order to optimize the benchmarks on Oracle SGBD.

[haplo602]

actual server benchmarks ?

Why was I expecting virtualisation or database workload benchmarks when I saw Oracle Linux in the title ?

ah yes, this is phoronix ...

[loll]

I'm most surprised by the difference between SL and CentOS.

Oracle probably applied some patches (so it's not completely equal to RHEL) in order to optimize the benchmarks on Oracle SGBD.

same remark.

Oracle kernel is different, seems that SL does the same ?
The CentOS kernel is it the one from centosplus repot ?

[DaemonFC]

Half the reason for the "UEK" kernel is because Red Hat obscures the particular patches they apply to thwart Oracle, and the other part is probably Oracle sacrificing a lot of workloads to boost performance with their own products.

Regardless of why Oracle's kernel handles common workloads so poorly. Ouch.

[CantankerousCuss]

Oracle Linux (UEK kernel) is intended for RDBMS server workloads

Be interesting to compare apples to apples with some I/O and RDBMS loads. Otherwise the tests risk discovering that a hammer makes a poor screwdriver.