MLPPP optimization

Nocturnal

Anyone here work with MLPPP WAN links across multiple platforms? I'm working on some issues dealing with throughput problems across WAN links using MLPPP. I've been expreimenting with MRRU tweaks and shutting off fragmentation on the CPE side. Anyone have experience with this stuff? I'd love to chat with you.

Also, I'd be interested in getting some real dialogue going on with MPLS and BGP if anyone's interested. I'm just starting to get into BGP and I have a rough overview of MPLS.

Let's get something going on here. I'm willing to do my homework and share if anyone else is interested.

tunerX

What kind of throughput problems are you having? Are you using a multilink interface? What exactly are you trying to accomplish?

Nocturnal

I get complaints from people running multilink accross two or three T1 lines claiming that their transfer rates are less than those running over a single T1 with HDLC encap. It is always with connections to non-Cisco edge routers. When I check the individual controllers I see no errors--I'm on the ISP side. The MRRU tweaks eliminate most of the bad fragments, but if I hammer the CPE WAN interface with oversized pings I can usually generate some, but that shouldn't be an issue because they should only be sending out what is agreed upon in the ppp negotiations, which is usually 1524.

I suspect it has something to do with the TCP connection between the end devices, but I have a hard time getting the information I need about the end devices and the overall architecture of the LANs.

Thanks for your interest!

tunerX

Is the customer running GRE or IPSec across the link? Depending on the options using GRE and IPSec can add around 60 bytes. With the MLP header and others you will be well above the 1524 limit.

This could cause problems but is kind of unlikely.

keenon

i think i'm doing something similar to what u have going with mlppp.. but i'm still working on getting the bugs out.

but in this i'm on the customer side

Nocturnal

TunerX: The tunneling is provided on our end from edge router to edge router, although I have worked with people who tunnel within our tunnels. It probably wouldn't hurt to ask, but I agree with you that it isn't very likely.

Keenon: I'm looking for the big tweak--the MRRU adjustments definitely get rid of most of the bad frags on my end, but I haven't had anyone try anything more than 1524--I think you can bump a Cisco up to 4500. Then there's the usual stuff like disabling CDP and multilink fragmentation. There's got to be something else. It really kills me not being able to cruise out to the customer end. Have you tried any debugs? I am forbidden to debug anything...for now.

BTW, is it possible to simulate a T1 connection by hooking two external CSUs together with a crossover? It seems like it might work, but then again that seems too easy.

tunerX

Yeah, you can simulate the T1. Just make sure that the CSUs are configured with the same linecode and framing, then connnect them with a crossover cable. You will either need to provide timing from the router and pass it through the link to the other router or inject timing into one CSU and pass it to both routers.

router---csu---crossover cable---csu---router
/
/
external timing

If you are using Cisco T1 modules with an integrated CSU/DSU then you can just use "clock source internal" and "clock source line".

The crossover cable is 1-4, and 2-5. This is for the Cisco modules or CSUs that have the RJ-48 plug.

1 4
2 5
3 N/A
4 1
5 2
6 N/A
7 N/A
8 N/A

Nocturnal

Wow...okay. I made my own T1 loopback plug just to get the WIC to stay up, but I only have one integrated unit. I think I might be able to find some used CSUs, though. We'll see. I just had to hear that it would work before I considered spending any money on more hardware.

Thanks!!