|Abstract:||Many studies have shown that the transmission con-
trol protocol (TCP) which is the major transport protocol in the
Internet today is finding it increasingly difficult to cope with the
growth of communication network capacities and applications.
TCP’s inability to properly utilize network links is one of the
problems. Besides, TCP takes a long time to achieve fairness
between flows. Many of the new modification of TCP inherit
these main problems of TCP.
Clean-slate protocols such as the explicit Congestion control
Protocol (XCP) which get congestion feedback from routers
can fully utilize the links and reduce queueing delays in the
path of the flows. But XCP, in addition to having many router
computation overheads also takes many rounds to allocate fair
shares to flows. To solve the drawbacks of XCP and TCP, we
have previously presented a network control protocol (NCP).
NCP allocates fair share to flows in one round resulting in
increased average file completion time (AFCT) as short flows are
not blocked by big file transfer flows. Although smaller than that
of XCP, NCP uses a 32 bit additional header in every packet to
carry a congestion feedback from the routers in the path of every
flow to each source of the flow. Apart from the inconvenience of
modifying the TCP/IP packet formats, the additional header may
accumulate to cause some increase in AFCT.
In this paper we present an efficient implementation scheme
(algorithm) of the Network Congestion Control (NCP) protocol
using Explicit Congestion Notification (ECN) bits (eNCP). Using
this implementation scheme, NCP does not need any additional
packet header, avoiding the per packet overhead. In addition
to the convenience of not having to change the TCP/IP packet
header format, numerical results show that a significant amount
of data overhead can be saved using eNCP. Not having to add
a layer can also make NCP easily deployable and backward