1) Dynamics of large (e.g. 20,000) flows of short duration (ie
10 pkts) through a bottleneck link.
1a) Generalization of 1) to see if the composition of x% bulk
streams and 1-x% small streams can be approximated by combining analysis
for steady state flows and flow startup.
1b) Interaction between RED+WFQ and highly aggregated traffic
2) Eager congestion control. Simulation of wart that eagerly
drops packets with same bottleneck link. Trick is in guessing how much
congestion there is.
3) Early Flow Discard... Combination of Tom's idea and mine wrt
edge-based dropping. When a packet is dropped, drop all future packets
until that sequence number is retransmitted. As an extra feature, send
back three duplicate ACK's for last ACK'd packet so sender doesn't have
to wait for RTT.
4) Combination of 2&3 with source quench model for interior
routers (e.g. when congestion found at interior send message to edge to
start dropping more).
5) ACK priority. Dropping other traffic in defference to ACKs.
What effect does this have on congested link? Interplay with ECN?
6) SYN priority (same as ACK)... how does this effect
probability of getting a connection under high loss?
- Stefan
> -----Original Message-----
> From: tom@emigrant [SMTP:tom@emigrant]
> Sent: Thursday, May 14, 1998 5:40 PM
> To: syn@cs
> Subject: tomorrow's meeting
>
>
> faculty club, in the conference room (different place than last time).
> I may be a few minutes late, so please start without me. Two things
> on the agenda: everyone's status, and brainstorming ideas for small
> simulation and measurement studies.
>
> tom