blob: 06567b0ad74f0cb39139de4b14df84d6a00c4ee3 (
plain) (
blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
|
- URGENT:
+ if 'client-not-ready', we do not ACK at all, and sender keeps
retransmitting again and again; would be good to do flow-control notification instead
of not ACKing that we got the data but are simply not ready for more!
+ Congestion/flow control (CHANNEL):
estimate max bandwidth using bursts and use to for CONGESTION CONTROL!
(and figure out how/where to use this!)
- HIGH: revisit handling of 'unbuffered' traffic! (CHANNEL/TUNNEL)
(need to push down through tunnel into connection selection);
At Tunnel-level, try to create connections that match channel
preferences (buffered/unbuffered) and select connections for
channel traffic that match channel preferences.
BUT: not sure this is ideal, discloses traffic type to
routers. We don't want that! (Maybe revise decision to do this?)
- HIGH: revisit handling of 'buffered' traffic: 4 is a rather small buffer; (CHANNEL)
maybe reserve more bits in 'options' to allow for buffer size control?
Or: maybe even better, calculated required buffer size based on latency
and throughput (and available memory)
- HIGH: if we receive BROKEN messages, cut down corresponding PATH (up to the
point of breakage) as well as connection/route (CORE)
- OPTIMIZATION: proper connection evaluation during connection management:
+ TUNNELS:
* consider quality of current connection set when deciding
how often to do maintenance
* interact with PEER to drive DHT GET/PUT operations based
on how much we like our connections
- OPTIMIZATION: optimize stopping/restarting DHT search to situations
where we actually need it (i.e. not if we have a direct connection,
or if we already have plenty of good short ones, or maybe even
to take a break if we have some connections and have searched a lot (?)) (PEER)
|
