[c-nsp] ppp multiclass

[Gary Roberton]

Hello All

This question was floated around a few days ago but I am still unsure
of a resolution.

We are using two 256k DSL links bonded together using ppp multilink
and running voice traffic across this.  The problem lies in the fact
that;

voice packet 1 travels physically along link 1
voice packet 2 travels physically along link 2
voice packet 3 travels physically along link 1

If anything happens to packet 2 the receiving router gets 1 and 3,
obviously not good.  We have therefore used the ppp multilink
multiclass statement.  This puts sequence numbers on the voice
packets/fragments and the receiving router will dynamically create a
buffer to reorder anything that is received out of order before
forwarding onto the LAN.  This should smooth out any variance in delay
between the two physical links.

The problem is that the receiving router waits up to 1000 ms for any
late fragments.  This is too long as it will introduce an unacceptable
delay to packets that are slightly late.  Shown here;

Virtual-Access5, bundle name is <removed>
Bundle up for 6d16h, 63/255 load, 2 receive classes, 2 transmit classes
Receive buffer limit 24384 bytes per class, frag timeout 1000 ms
Dialer interface is Dialer0
Receive Class 0:
  3/724 fragments/bytes in reassembly list
  1 lost fragments, 5397672 reordered
  0/0 discarded fragments/bytes, 0 lost received
  0xEDB5DD received sequence
Receive Class 1:
  0/0 fragments/bytes in reassembly list
  0 lost fragments, 0 reordered
  0/0 discarded fragments/bytes, 0 lost received
  0x0 received sequence
Transmit Class 0:
  0x5A981B sent sequence
Transmit Class 1:
  0x313F82 sent sequence
Member links: 2 (max not set, min not set)
  Vi3, since 6d16h, 640 weight, 632 frag size
  Vi4, since 6d16h, 640 weight, 632 frag size

Someone here suggested that we try the ppp multilink slippage command
but this sets a minimum time to wait for delayed packets.  Does anyone
know how we can reduce the time the receiving router will wait
(perhaps 50 ms) so that we can cover up slight differences between our
physical links without introducing too much delay.

Thanks in advance

Gary