>From testing I've done, ML-PPP re-ordering has a negligable impact on TCP
end-host re-ordering. I found that well over 50% of packets were re-ordered
at the far end of the ML-PPP link, and yet less than 1% were re-ordered at
the receiving host of a TCP session, with low latency and jitter.
You are right in using the 'no-fragment' option, I would also suggest you
use CEF, distributed if relevant.
Rgds,
Iwan
Steven W. Raymond said:
> Problem:
> User complains about TCP packets arriving out-of-order on the other end
> of a 4 T1 MLPPP bundle. Assertion is that the misordering is causing
> throughput problems when TCP detects the out of order MLPPP fragments.
>
> Issuing 'sh ppp multilink' the "reordered" stat is climbing like crazy.
> We're talking 300-1000 per second:
>
> router#sh ppp multilink
>
> Multilink2, bundle name is
> 36 lost fragments, 15371129 reordered, 0 unassigned
> 3 discarded, 21644 lost received, 125/255 load
> 0x375D0E received sequence, 0x45CAD4 sent sequence
> Member links: 4 active, 0 inactive (max not set, min not set)
> Serial5/1/0/27:0
> Serial3/1/0/17:0
> Serial11/1/0/22:0
> Serial10/1/0/28:0
> Multilink2, bundle name is
> 36 lost fragments, 15371695 reordered, 0 unassigned
> 3 discarded, 21644 lost received, 125/255 load
> 0x37633C received sequence, 0x45CD6C sent sequence
> Member links: 4 active, 0 inactive (max not set, min not set)
> Serial5/1/0/27:0
> Serial3/1/0/17:0
> Serial11/1/0/22:0
> Serial10/1/0/28:0
>
> Questions:
> 1) Does the "reordered" statistic suggest that TCP packets will/can
> arrive out-of-order?
> 2) Is there any way to improve this behavior? Do have "no ppp
> multilink fragmentation" configured on the multilink interface, but I'm
> not sure what it does exactly or even if it is relevant to this issue.
> 3) Are there better ways of load balancing across 4 T1s?
>
> Have previously load balanced using "ip load-sharing per-packet" &
> "per-destination" but have the understanding that this is an even worse
> method for potentially misordering packets.
>
> Cisco advises that some reordering is normal and okay, but it seems to
> me that at this rate it is too high and could lead to TCP packets out
> of order, causing less than full throughput.
>
> Any advice is appreciated.
>
> Thanks!
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:03 EDT