Struggling with qos and dropped calls

phoeneous

Trying to figure out why I keep getting random dropped calls...

Consider this scenario:

Say I have a small site with 10 voip phones, 2960, and 2901. No qos is done at the 2960. 2901 is connected to mpls at 1.5Mbps.

With this config:

policy-map WAN_QOS
class VOIP_RTP
priority percent 60
class VOIP_SIP
priority percent 10
class class-default
random-detect

Upon congestion, traffic matching class voip_rtp is guaranteed 921 kbps and traffic matching VOIP_SIP is guaranteed 153 kbps. The remaining traffic gets 470 kbps and is randomly dropped??

What would happen during congestion if I changed the config to this:

policy-map WAN_QOS
class VOIP_RTP
bandwidth percent 60
class VOIP_SIP
bandwidth percent 10
class class-default
random-detect

networker050184

Are you actually seeing drops on your equipment? That seems like it should be enough priority for the calls.

You don't need the priority for the signaling. It's not quite as delay sensitive as the voice.

Is this one a T-1 card or is it subrate? If so you are going to want nested policy maps.

If you change it to the bottom map what is going to happen is all traffic is going to get a certain amount of bandwidth. The problem is that the voice will not be scheduled to be sent first. So while it will have 60 % of the bandwidth it won't be the first 60% as it should get.

The first thing I would do is try to monitor the situation a little more closely. This could be a case of the traffic dropping on the carrier network and you chasing something you will never be able to fix. You need to find the exact place of traffic drop if you are ever going to get it fixed.

shodown

are the call's being dropped on net or off net?
Next at the location is it a PRI, or SIP trunk, or FXO or none?

PCHoldmann

As networker050184 said, the control traffic normally is just "bandwidth" rather than "priority." RTP should stay "priority."

You may want to look for egress drops at the hub site (drops on the carrier network towards yours.) Your carrier should be able to provide this for you. With multiple spoke sites, the hub circuit can sometimes be overrun.

Do you have QoS configured on the carrier?

How are you defining the classes?

If you run "show policy-map interface <interface>" do you see packets hitting the VoIP classes? For instance, here you can see 15 packets hitting the "ICMP" class:

R1#sh policy-map interface serial 1/0
Serial1/0

Service-policy output: OUT

Class-map: ICMP (match-all)
15 packets, 1560 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: protocol icmp
Queueing
Strict Priority
Output Queue: Conversation 264
Bandwidth 128 (kbps) Burst 3200 (Bytes)
(pkts matched/bytes matched) 15/1560
(total drops/bytes drops) 0/0

Class-map: class-default (match-any)
24 packets, 1792 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: any

phoeneous

Out for the next few days, will follow up soon but thanks for the feedback.

darkerz

Request the carrier pull 24-72 hour PM stats and put up some IP SLA and SNMPv3 standard stats, see if we're seeing a L1, L2 or L3 issue that would be obvious. Don't rule out over utilization nor line integrity, working with a large ISP/MSP, I promise you- Out of our 15,000 customers (businesses) I get 1-2 ever day that require a LEC or ILEC dispatch between the CO and Premise. That's from carriers like Sprint, Century Link and Covad/MP, even XO.

phoeneous

I think I've got it fixed. They didn't tell me that these were internal video calls having issues. In my class map for rtp, I wasn't matching dscp af41 traffic. Since I've added this, most of the issues have gone away. Will continue to monitor.

Oh, and ISP hasn't really been helpful. They say that their edge routers are matching dscp46 @512k and cos6 @ 512k.

shodown, what does on net and off net mean? And the hub site is connected to hq via mpls. No pri.

pcholdmann, yes traffic is matching both rtp and sip classes.

drkat

On-Net would be on-network calls, which is happening.. off-net would the PSTN is involved.

phoeneous

drkat wrote: »

On-Net would be on-network calls, which is happening.. off-net would the PSTN is involved.

Got it. Then yes, the calls are on net.

OT Question, if a call goes from branch to hq over mpls (and not a pri) how can I see the active call in session? When I run sh voice call stat or sh voice call sum, it only shows calls that are going out the serial interface connected to the pri.

drkat

Depends

Are you using CUCM? What does the topology look like? show voice call * is for voice-ports so that's why you're seeing the PRI stuff.

phoeneous

drkat wrote: »

Depends

Are you using CUCM? What does the topology look like? show voice call * is for voice-ports so that's why you're seeing the PRI stuff.

Yes, an 8.6 cluster. Using a 2911 as h.323 gateway.

drkat

from the gateway you should look at show h225 * commands