04-04-2012 07:30 AM - edited 03-04-2019 03:54 PM
I have an MLP with 2 T1s and having an issue with voice clipping. The policy for voice doesn't show any drops. I know that interleaving is recommended for < 768k links, but what about 3Mb MLP interfaces? Everything that I'm reading about MLP and interleaving doesn't stipulate a speed requirement.
As a side note, I tried to remove fair queueing from the serial interface and it crashed the router. I found bug CSCsd09892:
no fair-queue causing VIP crash | |
Symtom: On a router, a VIP crash may occur when "no fair-queue" is applied to interfaces with service-policy. Condition: This is only seen when the "no fair-queue" command is applied to an interface with service-policy. Impact: N/A Workaround: None |
It seems to affect 15.0, which I have 15.0(1)M6 on these routers. The only fixed versions seem to affect 12.x and not 15, but the problem exists in 15. The way that I'm reading this it looks like I should be able to remove the service policy from the MLP interface and then make the "no fair-queue" change and reapply the policy. Does that sound like it would work?
Thanks,
John
.
Solved! Go to Solution.
04-05-2012 07:04 AM
John,
Time to isolate the issue. Can you break the MLPPP and see if you have a faulty T1?
That's one of the reasons why I hate MLPPP because a faulty line can cause problems that are really hard to troubleshoot.
Let's try LFI first on both sites then breaking the MLPPP would be my next suggestion.
Have you had the site working before w/o problems?
04-05-2012 07:13 AM
Edison,
Unfortunately, I don't think I'm going to be able to break this but I did clear the counters the other day thinking that I may have errors on one of the links. The interface at the main site is still clean with 0 errors and this is the site that hears the clipping from the other site. The other site in question has not had their counters cleared which I just did, so I'm going to be watching that throughout the day to see if there's an issue. If I get errors I'll get in touch with the provider. One other thing that did help was finding that their phone system had autonegotiated on the switch side at 100/h and that was a major source of problems.
Thanks!
John
04-05-2012 07:40 AM
Good effort!
Don't concentrate just on the WAN link but the entire path between the phones to/from the call manager.
With a T1, you may not experience local errors but the provider may have a 'dirty' line.
04-05-2012 07:52 AM
I believe you hit the nail on the head though....I cleared out those counters on the "other" side router and sure enough I wasn't seeing input/output increment on one of the serial links. After all of those changes last night, one of the serial links went down this morning so all of the phone testing was done with a down circuit...lol....I have a ticket in with ATT and will retest when it comes back up.
04-05-2012 12:59 PM
Okay...no go. After the circuit came up, we can talk for a while but then it starts degrading and then clears itself up. I enabled lfi on the mlp interface and it seems to have helped, but I can't see a reason as to why it's choppy at times.
I have the following configuration and I'm thinking of changing it:
policy-map COS
class HIGH
priority percent 33
set ip dscp ef
class MEDIUM
set ip dscp af31
bandwidth remaining percent 60
service-policy BGP
class LOW
bandwidth remaining percent 30
random-detect
set ip dscp af21
class class-default
bandwidth remaining percent 10
random-detect
set ip dscp default
1. Should I not use WRED?
2. I'm thinking about changing my classes to use bandwidth percent instead of remaining
3. Should I change the percentage to the llq?
I finally saw a drop in the llq:
Class-map: HIGH (match-all)
1215489 packets, 187131628 bytes
5 minute offered rate 116000 bps, drop rate 0 bps
Match: access-group name HIGH
Priority: 33% (1013 kbps), burst bytes 25300, b/w exceed drops: 7
QoS Set
dscp ef
Packets marked 1215490
I could raise the percentage, but I'm not sure if that would help.
04-05-2012 01:27 PM
Keep in mind, when you allocate a high % for the PQ it will potentially starve other queues.
Once you get into the 40% range, I recommend upgrading the circuit.
1) WRED shouldn't interfere with voice traffic on this situation
2) Go with EF = 35% AF31 = 35% AF21 = 20% Default = 10%
3) addressed above...
04-05-2012 02:25 PM
Edison,
Thanks for the response again. I shouldn't be seeing clipping though according to what my policies are showing right? The drops happened when I ran WAN Killer to saturate the link. The voice quality was amazingly okay when the test was going, but it was after I was done with the test that I heard the static. At one point it was very bad, but neither side was saturated at that point.....really weird....
04-06-2012 02:59 AM
Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
Indeed "weird", and the fact it worked well during your stress test would appear to confirm that it can work as expected.
These MLPPP links are true leased links end-to-end between your sites, i.e. they do not connect to any kind of "cloud"? If so, all the other interface stats are clean too?
04-06-2012 03:29 AM
Joseph,
All interfaces across the 2 sites in question are clean. They do run to our provider into an mpls network. The LSP isn't in my control unfortunately. We pay for qos every month so they honor our tags and it's definitely getting to the other side, but there's no way of telling if they're experiencing congestion within their network which was the only other thought that I had.
Also, I still suspect the phone system though because the local contact called my yesterday. I don't have voip phones at my location, so when she has to call an outside line it has to pick up a pots line. She told me that she just got off the phone with someone at my location and it was popping/crackling while she was talking to them which is going over the normal phone line. I just wonder if maybe they have a line or two that could be bad in their PRI or if there's something in the phone system that can be done to see that.
John
04-06-2012 05:21 AM
Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
Ah, MPLS eh? Indeed it's difficult to tell wha't going on inside there. I've found when dealing with any WAN provider, you need to "trust but verify". They too, like the rest of us, occasionally make mistakes. The also sometimes have network issues they are unaware of until you find them for them. Lastly, when working with QoS across MPLS (I'm supposing you see this as L3 VPN), you need to fully understand how to leverage their QoS model(s) and issues that can arise even with their QoS implementation.
PS:
BTW, what's VoIP using for its codec? Reason I ask, something like G.711 can be a bit more "robust".
04-06-2012 06:25 AM
Joseph,
My understanding is that some sites (not sure which) are using g.729 and the some are using g.711. The provider gave us the af markings to use because they map them to the correct cos on their end. I could create an acl that matches on one address to one address and then match inbound on that marking matching the source to see if it's carried through the network.
Thanks!
John
04-06-2012 08:28 AM
Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
Unfortunately, what a provider spits out (for markings) doesn't indicate they are treating the marked traffic correctly. To verify correct provider treatment, I've found this can be done by sending various combinations of various class traffic, to verify what comes out is as expected. For example, if I send VoIP class traffic that's less or equal to the agreed subscription rate for that class, I should be able to overrun one or more other classes and still obtain expected performance for the VoIP traffic. If I don't get these results, provider needs to do some explaining. (Note - to do these kind of tests you need to know (control) end-to-end traffic rates, i.e. an active multipoint can be difficult to test. Additionally, these kinds of stress confirmation tests can easily be disruptive to other production traffic. Ideally, such testing only has known quantity of test traffic.)
PS:
I once spent 3 months arguing with a tier-1 MPLS international provider that one of our connections to their MPLS cloud didn't seem to be working 100% correctly. They swore up, down and sideways, nothing wrong with their network. Eventually, they found a line card with out-of-date firmware (for the card) was very occasionally corrupting packets when under high load. They updated the firmware; end-to-end behavior finally became as I expected.
On multiple occasions, with them and other tier-1 WAN cloud providers (frame-relay, ATM, MPLS, MetroE), encountered minor provider configuration errors such that, again, end-to-end performance was less than I expected. One time just one link of a bundle (forget whether it was MLPPP or ATM IMUX) was misconfigured. Difficult for me to identify, but once I did, they quickly saw the issue, unlike the firmware issue above. I.e. these type of errors normally required me finding the link that wasn't quite up to snuff, but when I did, the service provider would quickly "see" and correct the problem.
04-06-2012 08:45 AM
Thanks Joseph...it sounds like I have some more testing to do
Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: