[multipathtcp] LISA - draft-barik-mptcp-lisa

[Michael Welzl <[e-mail]>]

(chair hat on)

[ This email goes to the MPTCP and ICCRG lists as an announcement. When answering, please discuss this in the ICCRG list. ]


Dear all in MPTCP & ICCRG,

Last Wednesday, in MPTCP, Runa Barik presented draft-barik-mptcp-lisa (LISA - a Linked Slow-Start Algorithm). For more documentation & code, see:
http://heim.ifi.uio.no/runabk/lisa/
LISA was first presented by Runa in Yokohama, with feedback that was addressed (with results shown in this presentation).

(chair hat off, co-author hat on)

LISA is a simple algorithm that prevents a large burst of packets going out during in MPTCP in slow start in one particular round-trip (when all but the first subflows join). E.g. with 4 subflows, MPTCP starts out sending using IW10 on the first subflow, and then after 3 RTTs, 3 more subflows start at the same time, adding 3 * IW10 on top of the ongoing slow start. This short but massive burst is consistently bad when we share a bottleneck, and often LISA also makes things better when not sharing a bottleneck.

(co-author hat off, chair hat on)

LISA doesn’t fit the current charter of MPTCP, so we decided to get this reviewed and discussed in ICCRG, and possibly (if the updated charter doesn’t fit congestion control work) published via ICCRG too.

(both hats on)

We would greatly appreciate if people could look at this work and provide comments on the draft soon.

Thank you all in advance!

Cheers,
Michael