MOS and Latency Measurements in IP SLA

jswan

I've noticed that when the WAN link to a site is under heavy load, the latency measurement from the UDP jitter operation in IP SLA is often considerably higher (as in several hundred milliseconds higher) than the latency as measured by ICMP in NPM (this is the regular ICMP poll I'm talking about, not an ICMP echo operation in IP SLA). This makes no sense to me, because:

1. The UDP jitter operation is in the priority queue (to simulate voice traffic).
2. IP SLA operations are processed by the router at interrupt level, whereas ICMP replies are processed in the slow path (CEF vs process switching).

It seems like the ICMP measurement should almost always be higher. The extremely high latency measurements from IP SLA cause the MOS to dive, but users don't report call quality problems during that time.

Any ideas on what's going on here?

bob55

Hi,

I'm not sure if it's accurate to compare ICMP to IP SLA UDP jitter operations. Yes they should go hand in hand as in if ICMP is high then so should IP SLA UDP results. Now whether how big of a delta between the two would be hard to measure. IP SLA operations were design to be more accurate then ICMP due to the number of packets that it's sending every 10ms (default timer), and taking the delta between the two. Also it's more accurate because there's sending and receiving time stamps from the source and the destination operation target so that delay can be measured much more accurately.

MOS score gives a good general idea of sound quality but low score might just mean that you have degraded voice and not necessary jitter and dropped calls. There's also IOS variation as from one revision to the other they have changed metrics for different measurements. With that being said, I wouldn't worry about it to much and rather just try to get a baseline on ideal IPSLA UDP operations and go from there.

Robert