[systemd-devel] Excessive (virtual) memory usage of journald

[Holger Hans Peter Freyther]

On Sun, Jul 07, 2013 at 09:18:01AM +0200, Holger Hans Peter Freyther wrote:

Good Morning,

> > So unless there is an issue with my recording/replay I think that
> > besides my opinion that mapping a < 4MB file 65 times is ugly, it
> > also appears to be slower for the above workload in journald.
> 
> ping? any comments? what workload should benefit from the mmap cache?

I saw that you made some changes to journald that could benefit the
performance (cache cgroup root path, allocate larger blocks) and wonder
if you want or have re-visited your mmap cache.

In the trace/record of the append only case I showed that all allocated
Window.size were smaller WINDOW_SIZE (bloating the address space, making
finding real leaks more difficult and was a slow down in the benchmark).

Now with your "allocate larger blocks" how many windows do you end up
with for the minimal journal? What is the hit/miss ratio on these Windows?


cheers
	holger