Traffic Shaping / Policing Question

I must have a problem with my configuration. Thanks in advance for the help.

I simply have some TDMoIP boxes (T1 to ethernet) on each side of a slow WAN link (approx 24Mbps). On each end of the WAN link is a

2620XL router running 12.2.40a. These routers have FastEthernet interfaces facing the WAN link. The TDMoIP boxes use 3Mbps of constant bandwidth. The customer has run fine for a few years, but has started seeing drops associated with high loads on their link.

My reasoning in setting up the service policy was that I needed to reserve (priority) 3 Mbps of traffic for the TDMoIP and then throttle the rest to an ammount less than the capacity of the WAN link.

So I create 2 classes: VOIP with priority 3072 and class-default with shape average 15000000. (using a total of 18Mbps). This should be well under the link capacity (24Mbps) and I should not see congestion.

When I apply the service policy, I can see the counters in each class respond as expected. Total traffic seems limited to approximately 18Mbps. But the TDM boxes suffer greatly when the link is loaded. All interactive traffic is also thwarted big time under shaping or policing.

I have tried nesting the service policies as seen in some examples on Cisco's page, but even then, the problems did not subside.

I have tried policing instead of shaping with various burst values for both.

Here is my config (its the same on both routers save for the source and destination are swapped for classifying the Voice traffic):

class-map match-all VoiceRAD match access-group 157 ! ! policy-map RadioLink_pmap class VoiceRAD priority 3072 class class-default shape average 15000000 ! interface FastEthernet0/0 ip address 172.31.1.121 255.255.0.0 service-policy output RadioLink_pmap ! access-list 157 permit ip host 10.35.1.10 host 10.34.9.95 access-list 157 permit ip host 172.31.1.121 any

I added the router's own IP in the access-list since it was my means of configuring the router. Even after adding it to the priority class, I would routinely wait 60-90 seconds for keystrokes to appear.

Here is the output of the show service policy int:

cte2600#show policy-map int FastEthernet0/0

Service-policy output: RadioLink_pmap

Class-map: VoiceRAD (match-all) 4716486 packets, 424579155 bytes 5 minute offered rate 2950000 bps, drop rate 0 bps Match: access-group 157 Queueing Strict Priority Output Queue: Conversation 264 Bandwidth 3072 (kbps) Burst 76800 (Bytes) (pkts matched/bytes matched) 4444/496383 (total drops/bytes drops) 0/0

Class-map: class-default (match-any) 1258995 packets, 1746625441 bytes 5 minute offered rate 14188000 bps, drop rate 0 bps Match: any Traffic Shaping Target/Average Byte Sustain Excess Interval Increment Rate Limit bits/int bits/int (ms) (bytes) 15000000/15000000 93750 375000 375000 25 46875

Adapt Queue Packets Bytes Packets Bytes Shaping Active Depth Delayed Delayed Active - 0 1259057 1746711591 226821 324632790 no

This snapshot was taken shorly after load was removed, but you can see the offered 5-minute rate was under the threshold set, and also the 5- minute load on the voice traffic was as expected (3Mbps) and has a drop rate of 0. The only thing odd that I noticed was the line: (pkts matched/bytes matched) 4444/496383 under the Voice section - I dont know what it means by matches - all

4.7 million packets in that policy should be considered for the class.

I thought policing would be better - just drop anything over 15000000 bps, but it didnt seem to work either.

I am at a loss of what to do next. I've tried a bunch of different ways to configure the shaping/policing without success.

Thanks again.

Reply to
jere.cassidy
Loading thread data ...

What does the CPU look like during the peak traffic times? I would suspect that the 2620 doesn't have enough horse power for what you are trying to do.

Reply to
Scooby

Hide quoted text -

This was one of my fears. The CPU does appear to be hammered for short periods of time- spiking up to 99%/99% on the 5 second average. But generally, it was ranging in the 75-90% range. The CLI goes unresponsive, but I took that as more of a packet drop/delay problem than a CPU issue.

Is there a way to determine how much a 2620XL should be able to shape and/or police? again, I am working with a 24Mbps link trying to prioritize 3Mbps and limit all other data to 18Mbps. Is there a chart / whitepaper on Cisco's site?

They also have a 4607R on just one side of the link. I am not sure what type of switch is on the other side, but I'd expect a 2950 or some kind of workgroup switch. I may try to attack the traffic shaping/policing with those devices.

Thanks for the help!

-Jere

Reply to
jere.cassidy

The policing/shaping is not the problem. This router is just not strong enough to handle that volume of traffic. Do make sure that you have CEF enabled, but I still think you will need to replace the router to get the throughput you are looking for. That said, it probably is wise to move where you are doing the shaping. You want to handle this as close to the source as you can. You may need to find the setting that keeps the traffic on your router to a managable level. But, that means you are wasting bandwidth.

Cisco does have some white papers available. I don't have any links available - sorry. I have reviewed them before and they are worth looking at, just don't have them handy. The maximum throughput on routers is not rated on bandwidth, but on pps. The idea being that it takes about as much processing power for a small packet as it does for a big one. Try searching Cisco's site for pps throughput and your router model. You can do the math on the pps numbers to get the max bandwidth expected based on min/max packet sizes.

Hope that helps,

Jim

Reply to
Scooby

Thanks a lot Jim. I have located the pps numbers on the 2620XL and its 30000 pps. But that is straight from the data sheet and I am sure is the standard packet rate without any kind of shaping or policing. I will look arround for some papers that indicate how the PPS would be reduced by shaping/policing. I did notice that the backplane of the 4507R was 75million pps. So that device should have a lot more leeway with shaping/policing applied. If the switch on the remote side is an old 2950 or something, I'll just have to deal with that in another way.

Thanks again!

Reply to
jere.cassidy

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.