[rsyslog-notify] Forum Thread: Re: Dequeue Perfomance - (Mode 'edit_last_post')

[noreply at adiscon.com]

User: kamermans 
Forumlink: http://kb.monitorware.com/viewtopic.php?p=25473#p25473

Message: 
----------
Ok cool, that makes sense, so I'll set the
[b:3t2f6xh4]queue.size[/b:3t2f6xh4] to a reasonable estimate for my traffic
and memory (15GB RAM / 512b msg = ~30000, so I'll go with 20000).
Since I'm using DA mode, if my memory-based
[b:3t2f6xh4]LinkedList[/b:3t2f6xh4] queue fills up, the disk-assisted queue
continue to soak up messages until just a bit past my
[b:3t2f6xh4]queue.maxDiskSpace[/b:3t2f6xh4] (rsyslog always writes full
messages, even if they overflow the size limit).

This certainly makes sense to me.  I'll make these changes now and see how
it goes in the morning.

I'm assuming I'll still have performance problems, so perhaps I should add
multiple queues, one for each continent, region, datacenter, etc, and set
the [b:3t2f6xh4]queue.size[/b:3t2f6xh4] to 1/n, where n is the number of
partitioned queues, so they don't overflow the memory if things get busy in
all regions.  If I have 6 of these, my impression is that input and
dequeuing will be 6 times faster (assuming no CPU or I/O contention) since
they each have their own thread pools, can you confirm?