Database Reference
In-Depth Information
4.4.2 Alternative Transport Protocols
UDP, or the user datagram protocol,
68
is a thin transport protocol defined
on top of IP that provides de-multiplexing of application messages between
various applications running on the same host. In effect, UDP offers little
more than IP, which is addressing, routing, and best-effort packet delivery.
As such, UDP is perfectly suited for implementing various other transport
protocols. SCTP, or the stream control transmission protocol, is the latest
approved standard transport protocol.
78
Similar to TCP it provides reliable
transmission and uses the same congestion control mechanism; however, the
byte stream semantics have been replaced by message delivery. SCTP supports
multiple delivery modes (ordered, partially ordered, and unordered).
As the number of applications that require high-performance wide area
transfers has increased, so has the field of research in transport protocols. A
survey on the subject
40
classifies newly emerging protocols in three categories:
TCP variants, UDP based, and protocols requiring router support. A short
list of some of these protocols and their distinctive design ideas regarding
congestion control follows.
Scalable TCP
50
and HighSpeed TCP
28
use a similar congestion control de-
tection and avoidance mechanism as standard TCP but use a modified re-
sponse function to packet loss and acknowledgment. In H-TCP
57
the increase
function
a
depends on the time elapsed since the last congestion event, and the
decrease function
b
depends both on throughput and ratio between minimum
and maximum delay experienced by a source.
TCP Vegas
15
introduced a congestion avoidance algorithm based on mea-
suring time delays of individual packets. The congestion window is updated
based on measured differences between packet round-trip transit times and
ideal network round-trip time of the uncongested network. The same approach
is used by a newer protocol, FAST TCP.
84
Compound TCP
48
combines these
two approaches (delay based and loss based) into a single algorithm. Its con-
gestion window has two components: a loss-based window, which is computed
as in standard TCP, and a delay-based computed window added on top of it.
BI-TCP
88
introduces a new technique for determining the available window
size called
binary search increase
. In response to acknowledgment BI-TCP
increases its window to the midpoint between the current window size and
the maximum known size if this increase does not exceed a certain threshold,
or with the threshold value if it does (additive increase). When the window
exceeds the maximum, the available bandwidth is probed using a symmetric
response function. The response function was updated in CUBIC
72
from the
binary search logarithmic to a strictly cubic function.
TCP Westwood and the updated version TCP Westwood+
63
use a tech-
nique called
adaptive decrease
, which in response to a congestion event tries
to guess the available network bandwidth and sets the windows size to the
guessed value. The response to acknowledgments remains that of standard
TCP.
