07-17-2008 12:15 PM - edited 03-03-2019 10:47 PM
Are there any recommendations as to whether multiple T1s (private point-to-point leased lines) should come from the same service provider or whether the buffering and reordering capabilites of MLPPP compensate for the difference in latencies that one usually gets with circuits from different providers. Having different providers does offer some protection against all of your T1s going down if the provider has a problem, but I'm not sure how the difference in latencies might affect MLPPP.
07-17-2008 12:19 PM
Hi, MLPPP has buffering to tolerate differences in latency, beside T1 circuits terminating to the same places will be pretty much the same overall.
Please rate post if it helps!
07-17-2008 12:48 PM
Although in theory you could use 2 providers with mlppp you have to be very careful. MLPPP ensures that packets to do get delivered out of order. To do this it holds the packets in memory. The risk here is that if too many packets arrive out of order because there is a large difference in the latency the router over runs the buffer and drop them.
A few ms difference should work but you may need to adjust the buffer size.
When I have ordered additional circuits even from the same provider they many times would replace all the circuits so they could guarantee the same latency. The difference can be very significant even though the lines in theory go between the same cities.
One of the arguments the providers will make as to why you should purchase both circuits from them is that they can guarantee a diverse path where if you use 2 providers they may share infrastructure in the path. Not sure I buy that since I have had both circuits fail and they were suppose to be diverse.
You could use 2 providers and either load balance by session or you could load balance by packet and let the pc's deal with out of order.
07-18-2008 05:29 AM
Thank you. We used to load balance by session, but once we started deploying VoIP, the recommendation was to use MLPPP. We already had T1s from two different providers to the remote site, so we tried it. We do see lots of reordered packets - about 30%. So, we will most likely get multiple T1s from the same provider for the next remote office to see how that works out. Is there any way to determine if the router has overrun the buffer and dropped packets?
07-18-2008 05:40 AM
It produces messages in the log if it gets real bad but show ppp multilink will even show how many packets it has reordered and dropped
07-18-2008 05:48 AM
What is the actual delay difference?
Note that 30% packets reorder doesn't necessarily mean there is any problem in MLPPP and can happen even with perfectly equal delay links, just consider the following scenario:
One larger packet is received on LAN interface, followed by a smaller one.
In absence of fragmentation (a perfectly valid option in MLPPP), router will transmit first packet on the the first link, and the second one on the second link.
However, the second and smallest packet will be completely received as first by the receiving router. So, to avoid generating Out of Order packets (A Very Bad Thing), receiving router will buffer the smaller packet, wait for the larger one to complete, then send them out in the proper sequence.
See also:
http://www.cisco.com/en/US/prod/collateral/modules/ps2033/prod_white_paper0900aecd8056d3cb.html
please rate post if it helps!
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: