[c-nsp] t-1 load balancing options to deal with apparent congestion

[Jeff Chan]

We have four parallel t-1s to a remote site.  Currently all four
interfaces are numbered to the same /30 and the load balancing is
the default per-destination, with CEF enabled on both ends.  One
router is a 7513/RSP4/PA-MC2T3+ on 12.3(13) and the other is a
3620/NM-1E2W/WIC-1DSU-T1 running 12.1(20).  All interfaces show
no errors and appear 100% clean.

Ping times across the WAN run around 2 ms most of the time, but
some occasionally go to 100 or 200 ms.  It's possibly congestion
of a particular T-1.  We're not using MLPPP so the maximum any
given connection will be able to use is about the 1.5 megabit
limit imposed by each T-1.

Are there other common sources of latency or congestion we should
be checking for?  What alternative configs should we consider,
particularly if we don't want to run MLPPP or per-packet load
balancing?  How does one diagnose sources of latency in such an
environment?

Cheers,

Jeff C.
-- 
Jeff Chan
mailto:cisco-nsp at jeffchan.com
http://www.supranet.net/