Hi Ged On Wed, 2011-10-26 at 08:35 +0100, G.W. Haywood wrote:
> ... iptables is perhaps a > crude tool for this sort of work and you haven't given us much else to > go on. Which operating systems do you use? On what hardware are they > running? Apart from running Smokeping, what else are they doing? Can > you describe the network for us? Is it entirely under your control? > What continuous data throughput rate(s) can it sustain? How many hops > are there between endpoints? What are the switches and routers? Are > they doing any kind of traffic shaping? What are the typical round > trip times? Have you reduced the network to a bare minimum set of > harware and repeated the tests? Some time ago I had buggy hardware > issues which limited the round trip times given by 'ping' to integer > multiples of 4ms - have you checked that the interval timers in the > smokeping machines are reliable? Have you attempted to use something > like Wireshark to monitor all the traffic on the network? Have you > tried inserting a router in the packet path which does nothing other > than time the packets? Are you running ntp on all machines involved > so that you can make (more or less:) absolute time measurements? The network is put together with Cisco switches and 1 or 10G links, the average round trip times would be under 10ms and we have sufficient bandwidth so no shaping or aggressive QoS at this point. I didn't concentrate on the network set up and conditions because the measurements are made from the same host to the same destination by the same software :) The accuracy of the measurements is a separate issue as far as it produces similar 'fuzziness' over time. The graphs can be seen here: http://www.mediafire.com/imgbnc.php/7a21ceb2cafbe64c4236189b62f458c007e595be963aa1a4df7a8fc0cdf55e456g.jpg The above shows the initial result to switch1, which changed (after adding a further probe to switch2) to a more or less flat line and then the reproduction. http://www.mediafire.com/imgbnc.php/4483d0cf3da608c48c2ac97ea9f703e1bf7ca3edfd0210fefbbe8434941873616g.jpg I added a separate probe to switch1 with identical settings, but offset by 30% so it runs shortly afterwards without overlapping. This was flat at all times regardless of the changes in smokeping configuration. The config files: Probes - http://www.mediafire.com/file/4rxpbuk79dp3y9d/Probes Targets - http://www.mediafire.com/file/w6abkbt7936kxfw/Targets > > > 2. Separate scaling issue follows from FPing being limited to sending > > pings at 1ms interval > > There will always be some limit, what's the question? :) Is there a way to avoid/circumvent the limit. In our case I will go over the limit of 150 probes in about a year, at a guess. That is if we are happy with basing our statistics on probes active only 10% of the time. As a comparison, the Cisco IP SLA udp-jitter probe runs with 1000 pkts over 20s every 60s by default - so 33% of the time. Nik _______________________________________________ smokeping-users mailing list [email protected] https://lists.oetiker.ch/cgi-bin/listinfo/smokeping-users
