Side_1_360

Some experiments show that RIO may offer bet-
ter results than WRED when parameters are cor-
rectly set. However, a number of additional
parameters are to be given in RIO, such as those
of IN priority. These are influenced by the given
scenario. Therefore, WRED may be preferred by
some when robustness is required.

Estimating the better combinations may well be
a challenge in itself, considering the number of
parameters that may be asked for and the
dynamics in the traffic flows.

2.2.2 Explicit Congestion Notification
Explicit Congestion Notification (ECN) has
been suggested as one way of tackling conges-
tion handling. Two bits in the IP header are
used: Congestion experienced (CE) – bit 7 of the
ToS field, and ECN capable transport (ECT) –
bit 6 of the ToS field. For IPv6 the correspond-
ing bits in the traffic class field are used. The
ECT bit is set by a sender able to react on indi-
cation of congestion by ECN.

In addition TCP has to be modified in order to
carry the information back to the sender. TCP
can be said to treat the network as a black box,
only reacting on lost packets without any further
insight into the situation in the network. This
might lead to low utilisation of the network.
Therefore, active queue management, ref. [RFC
2309] has been promoted to avoid some of the
drawbacks of packet dropping when queues are
(almost) full. Random Early Detection (RED)
is one example of an active queue management
mechanism.

The ordinary TCP algorithm will not assist app-
lications that are sensitive on delays (and pack-
ets arriving too late). Some algorithms, like

ECN, could be introduced to allow the source to

adjust its behaviour without experiencing too

low throughput (due to packet loss) or long

delays. Then the active queue management

could rather set the CE bit than drop the packet

when congestion is building up. This allows the

receiver to get the packet, avoiding retransmis-

sion, at the same time as a congestion indication

is conveyed. However, for IPv4 the header

checksum has to be updated when the CE bit is

set. This can be done incrementally as described

in [ID_ecn].

When a sender gets information on congestion

by ECN, it is supposed to behave as if a packet

was dropped. One argument for this is to provide

fairness compared to non-ECN systems. Thus a

router, in case a congestion threshold is ex-

ceeded, may drop a packet when the ECT bit is

not set and set the CE bit in packets where the

ECT bit is set.

Figure 8 RIO algorithm

Figure 9 Illustration of

ECN-related messages

Discarding Probability

Average

Queue Level

pmax

(Out)

pmax

(In)

p = 1

OUT IN

Minth(In) Maxth(In)

Minth(total) Maxth(total)

sender IP router receiver

congestion

IP(ECT bit set, TCP....)

IP(TCP_SYN-ACK(ECE flag set))

IP(ECT bit set, TCP....)

IP(TCP_SYN-ACK(ECE flag set, CWR flag not set))

IP(TCP_SYN(ECE flag set, CWR flag set))

IP(TCP_SYN-ACK(ECE flag set))

IP(ECT bit set, CE bit set, TCP....)

IP(ECT bit set, TCP(CWR flag set, ....)

IP(ECT bit set, TCP(CWR flag set, ....)

IP(TCP_SYN-ACK(ECE flag cleared))