[smokeping-users] need a performance recommendation

[Tobias Oetiker]

Hi Jeff,

Yesterday Jeff Williams wrote:

> I need recommendations to increase performance on our smokeping server. We
> are running smokeping 2.2.3 with rrdtool-1.3beta1 on a Dual Xeon 2ghz
> machine with 2g of ram. I'm not sure of the hard drive specs, but I believe
> they are at least 10K rpm Serial Attached SCSI.
>
> We currently have 6200+ ip addresses to ping, with that number increasing
> every few days as we add stores, and we would like to ping them 15-20 times,
> if possible, every 5 minutes. When I initially set this up, I created a
> number of different FPing probes with a script that creates the smokeping
> config file, dividing the ip addresses up into the different probes somewhat
> evenly. So, my questions are these:

> 1. Since each smokeping process and corresponding fping process take up
> their own share of system resources (disk I/O, cpu, memory), am I better off
> keeping the number of probes fairly low (< 5) or am I better off creating
> another probe each time there are another couple hundred ip addresses?

my guess is that fping will take some time to do all the pinging in
such a large setup the same is true for the updating of the rrd
files, so if you have multiple probes with forking, NOT doing the
same thing at the same time, chances are that you can utilize the
resources of your system better

> 2. Do my rrd settings affect whether the probes finish in their allotted
> time or not? By settings I mean the AVG, MIN, and MAX hb, steps and total
> settings.

the number of RRAs affects performance, the size does not ...

> 3. I am currently using an offset of 0% for each probe, because I don't care
> about how utilized the network becomes at this point, as well as a
> mininterval and hostinterval of 0.001. I also have concurrentprobes set to
> yes. Are those settings reasonable and do I need to be looking at other
> parameters?

as I said things should be spaced out across the update interval
for best performance

> 4. Is it possible for us to achieve the goal of 15-20 pings per ip address
> every 5 minutes with this hardware?

I would asume so, but since you do have the hardware I would
suggest you just try :-)

cheers
tobi

-- 
Tobi Oetiker, OETIKER+PARTNER AG, Aarweg 15 CH-4600 Olten
http://it.oetiker.ch tobi at oetiker.ch ++41 62 213 9902