On Wed, 2014-10-29 at 15:06 -0400, David Miller wrote:

However in the longer term I'd say that this value, if it is to have a
limit, then such a limit should probably be scaled based upon the
window size.

Yuchung and othres are working on a new way to handle reorders (RACK),
and should present the concept in next IETF meeting.

A linux patch should follow shortly.

High level idea is :

Decide when and what to retransmit based on the timing, instead of
sequence, relationships. This covers both original or retransmitted
packets.

On dupacks, wait a fraction of RTT before the repair process to both
allow reordering and relieve the network

Thanks