[hobbit] Conn time discrepancy

[Henrik Størner]

In <4ACCF0D4.3060307 at doublesparks.net> Alan Sparks <asparks at doublesparks.net> writes:

>wshart at hormel.com wrote:
>> From Xymon page :
>> <host ip> is alive (44 ms)
>>
>> From command line on Xymon :
>>
>> <host ip> is alive (0.13 ms)

>This is a known issue with the way hobbitping works, compared to fping. 
>If you want realistic ping times, you need to use fping.

I mostly agree.

One reason for this is that Xymon's ping-function is meant to test whether
the host is on the network; it isn't meant (really) to test how fast the
network connection is to the host. So Xymon runs loads of ping-tests in
parallel, and can easily get a much higher round-trip time than a ping
of a single host, running on an otherwise idle system. Of course - 
Xymon will notice if it takes 10 ms or 20 seconds to ping a host, but
whether it is 2 or 50 milliseconds really should not be cause for
concern. (E.g. if the ARP entry for the IP address is in the ARP cache
it will be some milliseconds faster then when it isn't, so the responsetime
depends on a "random" timeout in the OS kernel and the size of your ARP cache).

And hobbitping is even more aggressive with respect to running ping's
in parallel than fping.

I had a curious experience the other day. We've been upgrading the hardware
used for network tests from a (very) old Sun box to a somewhat newer and
more powerful Intel system. A couple of weeks later one of our account
managers came by; they were trying to figure out why the ping-times 
they could see in Xymon had suddenly become so much better. They hadn't
changed anything on the hosts or the network - but the Xymon server was
handling the ping tests much faster.

Performance testing is difficult.


Regards,
Henrik