WAN latency

jonathancert · ‎06-11-2008

I have a WAN link with high latency, usually around 950ms on my 'ping' replies. I'm using EIGRP but my neighbor adjacency drops of occasionally. Would using static routes solve this problem? My only concern is that I have two links connecting to the site, without a dynamic route will I be able to keep redundancy?

Jon Marshall · ‎06-11-2008

Jonathan

Rather than revert to static routes you may want to consider increasing the hello times on the routers for EIGRP.

http://www.cisco.com/en/US/tech/tk365/technologies_white_paper09186a0080094cb7.shtml#neighbordiscovery

You could use static routes if you wanted to. Are both links being used all the time or is one active and one passive - you can use a floating static for the passive link.

Jon

jamrjor · ‎06-11-2008

Thanks, I didn't think about using a floating static. Both links will not be used at the same time.

Joseph W. Doherty · ‎06-11-2008

950ms latency!? What are the properties and utilization of these links? What type of queue management (e.g. FIFO, WFQ, etc.)?

Reason I ask, I wonder if EIGRP packets are being overly delayed.

jonathancert · ‎06-12-2008

I'm using FIFO. What type of queue do you recommend? Yes, the latency is high, nothing i can do about that. We are talking modem, ATM's and satellite links. Shooting data across the world.

Joseph W. Doherty · ‎06-13-2008

Ah, satellite; certainly explains high latency.

If supported, and if the platform can provide it without overtaxing itself, I prefer FQ or WFQ to FIFO to minimize one (or a couple) of bandwidth heavy flows from overly delaying or causing synchronous tails drops with other traffic. It might also help with EIGRP.

jonathancert · ‎06-13-2008

Thanks, I will give it try.

mpatus · ‎06-27-2008

You may also try using LLQ and put your EIGRP traffic in the PQ. That should keep your EIGRP adjacencies up.

jonathancert · ‎06-30-2008

I'm not familiar with alot of queuing methods. Is this a difficult setup and does it need to be implemented on the local and remote router.

Joseph W. Doherty · ‎06-30-2008

Difficulty depends on how complex the queuing configuration. If you don't have a form of FQ, try it first.

If the interface is a serial of 2 Mbps or less, try:

interface serial #

fair-queue

If the interface is faster than 2 Mbps, try CBWFQ:

policy-map CBWFQ

interface FastEthernet #

service-policy output CBWFQ

If the interface is faster than the capacity of the link, "shape" to the capacity, e.g.

interface FastEthernet #

traffic-shape rate ### (in bps)

Placement is on both ends outbound.

jonathancert · ‎07-01-2008

Wow, everyone is giving me some great ideas. I plan on trying WRED combined with traffic-shaping this weekend and monitoring the status for any improvement or degrade.

Mark Yeates · ‎06-30-2008

As a former tech controller there isn't much you can do to reduce the latency over a satellite link. Queuing methods will only help if the link is saturated, and will provide little help overall in your current situation.

Mark

jonathancert · ‎06-30-2008

Go Waterwalker! I have given up on the latency issue, can't do anything about the transmission medias in use. The link is also saturated, I using Netflow to monitor it and sometimes I get burst at 110%. Working on getting the bandwidth increased but until that red tape gets conquered, maybe queuing will keep things stable.

Mark Yeates · ‎07-01-2008

I would definitely recommend a good queuing strategy to help with congestion. Just remember that it won't replace an upgrade in bandwidth, but like you said keep things more stable and it should help. Good Luck!

Mark