[rrd-developers] [rrd] Re: rrdcached performance with >200k nodes
Florian Forster
rrdtool at nospam.verplant.org
Wed Jan 13 11:33:22 CET 2010
Hi,
On Wed, Jan 13, 2010 at 11:08:44AM +0100, Mirek Lauš wrote:
> 31,901,245,144 ???:strcmp'2 [/lib/libc.so.7]
there are only three places where “strcmp” is used:
* When initializing the journaling code. Should be run at startup only.
* When checking for permission. This is called at least as often as
“strcasecmp”, so this is probably not the problem either.
* When looking up a node in the balanced binary search tree. This is
most likely the place where all those invocations come from.
I'd say, if looking up stuff from a balanced search tree is the most
time-consuming operation, all the rest is pretty awesome efficient.
> Write rate is very low, disk IO is almost zero.
> CPU load done by rrdcached gets very high (100-200%).
How exactly did you measure those numbers? Can you provide us with
absolute numbers, especially IO-operations per second? Disk throughput
is not a relevant measure because disk seeks are the limiting factor.
Are you sure the process is using the CPU for calculating stuff rather
than just waiting for IO? In a typical “the server is as busy as can be”
scenario you'd have 100% (times the thread count, the value specified
with `-t') percent CPU usage with ca. 10% system, 5% user and 85%
IO-wait.
By the way, the limiting factor is (or, should ideally be) physical
memory. How big is RRDCacheD's resident segment size (RSS)?
Regards,
—octo
--
Florian octo Forster
Hacker in training
GnuPG: 0x91523C3D
http://verplant.org/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.oetiker.ch/pipermail/rrd-developers/attachments/20100113/f49c6cd3/attachment.pgp
More information about the rrd-developers
mailing list