Hi Martin, On Tue, Jul 31, 2018 at 09:31:42AM +0200, Martin Drescher wrote: > I'm running a bunch of HTTP servers, most of them running a RHEL > 6, which is a kernel version 2.6. All patches applied, so that > Meltdown and Spectre stuff should be included. Some of the servers > where migrated to a Debian 9 with a kernel version 4.9, also all > patches applied. What I see is a tremendous difference in > /proc/loadavg. It is like 3 ~ 4 in kernel 2.6 and like 15 ~ 20 in > 4.9. It is not a real problem, that Debian is running well.
It is likely that the kernel's idea of what is a "runnable" process has changed since the vintage of the Red Hat kernel, compared to the kernel which Debian stable uses. The Debian kernel will be much closer to upstream. The RHEL kernel will be much older with backports only of security issues and serious bugs. Tweaking of load average stats probably did not make the grade for a backport. Here is some more information about how Linux load average is calculated. This may help you to narrow down which processes are being considered differently between the two kernels. <http://www.brendangregg.com/blog/2017-08-08/linux-load-averages.html> Cheers, Andy -- https://bitfolk.com/ -- No-nonsense VPS hosting