IP SLA accuracy

mszeftawy · ‎10-20-2009

Hi All

I have already implemented an IP SLA probe on my network because I need it badly to measure the voice performance (jitter and delay and packet lossâ€¦).

I used UDP jitter operation with the below configuration to measure the voice performance End-to-End.

I am using 12.4(25a) on 3845 platform, there are no well known bugs on this IOS that can impact the accuracy of my testing probe.

Now I have 2 questions

To which level I should trust the accuracy of this testing probe knowing that I have around 150Mbs of voice traffic in this direction?

Is there any need to increase the amount of traffic sent by the probe knowing that in such a case I may lose the regularity between the probe traffic and my actual voice traffic?

Are there any way to increase the below sample to simulate around 10 or 20 G711 voice calls knowing that my current CPU utilization does not exceed 10%(don't have CPU issues)?

Note: I don't care about the BW consumption all I need is to get the most accurate result I could. Knowing that I am testing STM-1s and STM-4s that carries a large amount of traffic?

ip sla monitor 1

Type jitter dest-ipaddr X.X.X.X dest-port 16386 source-ipaddr Y.Y.Y.Y codec g711alaw

Timeout 500

Frequency 21

Giuseppe Larosa · ‎10-21-2009

Hello Mohamed,

>> To which level I should trust the accuracy of this testing probe knowing that I have around 150Mbs of voice traffic in this direction?

the idea behind active monitoring is to create ad hoc traffic that follows the same path and is treated in the same way as all other VoIP flows or conversations by network devices in the middle.

For sure your STM-4 are not terminated on the C3845. you may have an STM-1 link but on the C3845 but it cannot handle 100 Mbps of Voip traffic with only 10% average cpu usage.

if C3845 is not on the WAN path your IP SLA should be treated as any other VoIP flow by other network devices.

The question may be : what can be a sampling ratio between N voip flows and IP SLAs?

I know that call manager clusters have auxiliary servers called CUOM CUSM that can provide reports of effective Voip calls.

We have them deployed and you can get reports on quality of Voip calls.

this uses the control part RTCP of RTP protocol that provides statistics for each real call collected by phone endpoints.

that can be a more meaningful monitoring tool that can say x percent of calls have had a service level less then this threshold.

But you may be a VoIP carrier that doesn't manage endpoints' VOIP PBXs

Hope to help

Giuseppe

mszeftawy · ‎10-21-2009

Hi Giuslar

Yes you are write the C3845 does not have the STM4 or even the STM1 interfaces, I am using it as shadow router connected just beside my voice equipment (soft switch-media gateway) and the remote responder router is also dedicated for SLA and connected just like the remote MGW, the network is for a mobile operator and we have a very large amount of traffic.

And as you can see that the configured probe simulates only one voice call, and I need to know is this single call(50 PPS) is enough to provide an accurate result to measure a network performance that carries a large amount of voice traffic or I need to modify my configuration.

Thanks