08-09-2007 05:02 AM - edited 03-03-2019 06:15 PM
Hi
The link Latency (Multilink with 2 x E1) normaly is 160ms, but some times occurs a latency peaks (1000ms)even with low Bandwidth Consumption (20%).Anybody knows the problem origin ? TKS
Solved! Go to Solution.
08-09-2007 05:38 AM
Also please remember, that concord (and other tools) measures bandwidth utilization that is average over time (like 5 minutes). But router/switch ALWAYS sends packets at maximum speed. I.E. The link can have two states: Either 0% utilized, or 100% utilized, but over time the results average out. If latency test is done in a time of high peak (e.g. 20 seconds in those 5 minutes the link is almost constantly loaded to 100%) then your queues and tx-rings will fill and you will have latency peak at that time, since latency is measured at certain instances.
Therefore there can be occasional peaks.
08-09-2007 05:05 AM
How are you testing this? Pinging directly between routers?
This is actually not a very reliable test of a links latency, as this ping is serviced by the processors on both routers. Not only does this take a different switching path than traffic transiting both switches (ordinary traffic is CEF switched), but responding to pings is the lowest priority process on the CPU.
When you see ping spikes when pinging between routers, it is more likely that the router was busy doing something more important at that instant, like recalculating a routing protocol, or servicing a burst of more important packets.
If you want to get a real latency test, ping between PC's connected to either end of the routers. You should see less spikes that way.
08-09-2007 05:10 AM
I verifed the latency and bandwidth with Concord eHealth 5.7 . Tks for your help!
08-09-2007 05:38 AM
Also please remember, that concord (and other tools) measures bandwidth utilization that is average over time (like 5 minutes). But router/switch ALWAYS sends packets at maximum speed. I.E. The link can have two states: Either 0% utilized, or 100% utilized, but over time the results average out. If latency test is done in a time of high peak (e.g. 20 seconds in those 5 minutes the link is almost constantly loaded to 100%) then your queues and tx-rings will fill and you will have latency peak at that time, since latency is measured at certain instances.
Therefore there can be occasional peaks.
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide