Database Reference
In-Depth Information
A different direction is represented by protocols that do not take a window-
based approach and thus indirectly control the data transmission rate, but
rather use a rate-based mechanism directly. These protocols use a fixed or
variable data transmission rate and no explicit congestion window.
Reliable Blast UDP, or RBUDP,
39
is a rate-based protocol that performs
reliable data transmission on top of UDP, and the data transmission rate is
specified by the user. Its successor, LambdaStream,
87
automatically modifies
the data transmission rate in case of congestion or when available bandwidth is
detected. Congestion is detected when the data-receiving rate is smaller than
the data-transmission rate, and detection is based on measuring inter-packet
delays at both the sender and the receiver.
UDT
36
also uses a rate-based congestion control algorithm. On acknowl-
edgments, each RTT rate is modified. The increase function depends on the
estimated link bandwidth, which is probed periodically. On a negative feed-
back the rate is decreased by a constant factor. UDT also supports a messaging
mode with optional ordering and reliability parameters.
The Group Transport Protocol, or GTP,
86
is a rate-based protocol design
that uses receiver-based flow management. Its design considers the problem
of multiple senders-single receiver flows explicitly. GTP uses flow statistics to
estimate the capacity of each flow and allocates bandwidth to each according
to the max-min fairness criteria.
The eXplicit Control Protocol (XCP)
49
is a router-supported transport pro-
tocol. Based on the observation that packet loss is not a reliable indication
of congestion, XCP uses feedback from the network (routers) in order for the
sender to correctly determine the degree of congestion in the network.
We should also note that with the advent of wide area optical network
links, high-speed protocols for wide area transport are being researched and
deployed outside the TCP/IP world. For example InfiniBand,
*
a high-speed
interconnect protocol originally designed for local area networks, can be ex-
panded to wide area networks using specialized hardware (Longbow by Ob-
sidian Research
†
). Fibre Channel extension to wide area networks was also
demonstrated by ADVA.
‡
Fibre Channel is an interconnect technology origi-
nally designed for local area networks.
$
4.4.3 Current Status and Options
Link utilization for dedicated or long-distance connections and fairness be-
tween concurrent transfers are two of the many possible metrics that can be
used to evaluate a transport protocol. There is no clear consensus, but there
*
http://www.infinibandta.org/specs/. Accessed July 16, 2009.
†
http://www.obsidianresearch.com/. Accessed July 16, 2009.
‡
http://www.advaoptical.com/. Accessed July 16, 2009.
$
http://www.fibrechannel.org/technology/overview.html. Accessed July 16, 2009.
