Announced at the XBIZ LA Digital Media Conference, FastTCP (www.fasttcp.com) is a MojoHost-backed rework of the TCP stack designed to deliver “The Internet, Faster.”
According to Wikipedia, the Transmission Control Protocol (TCP) uses a network congestion avoidance algorithm that includes various aspects of an additive increase / multiplicative decrease (AIMD) scheme, with other schemes such as slow-start, in order to achieve congestion avoidance. The resource goes on to state that the TCP congestion avoidance algorithm is the primary basis for congestion control on the Internet.
FastTCP dramatically boosts website speeds by providing a more efficient and intelligent TCP stack implementation that is compatible with existing infrastructures.
TCP was originally developed before the advent of broadband Internet, however, and has thus suffered from inherent limitations in its data throughput — until now.
“In normal conversation, when people discuss TCP they refer to a commonly used implementation, TCP Reno, or one of its variants,” FastTCP VP of Technology, Christopher O’Connell, told XBIZ. “Despite the almost universal adoption of Reno based variants, TCP exists as an abstract specification for the ‘language’ with which systems communicate.”
O’Connell says that is because TCP is a data interchange specification, it is possible to implement a superior algorithm that continues to meet the technical requirements of TCP.
Here, the algorithm is “Fast,” an apt name as it boosts speeds more than 150 percent.
“The TCP specification, in general, has one missing piece — intermediate routers cannot explicitly tell the origin that data has been delayed due to network congestion,” O’Connell explains. “Instead, routers delay data until their queue fills up, at which point packets are dropped.”
Typical TCP implementations use a sole binary digit (0 or 1), to track packet arrivals.
“As a result, current TCP implementations send data at maximum speed until data is lost, at which point they must re-send the lost data, and radically slow their sending rate until the connection recovers,” O’Connell added. “In practice, however, existing TCP implementations rarely run at full speed and spend most of the time in some form of speed limited state.”
FastTCP gets around this limitation by keeping track of the time that it takes packets to arrive and then computing a delay factor.
“As intermediate queues fill, the delay factor increases,” O’Connell notes. “By using the many bits of information contained in the delay calculation, FastTCP optimizes the connection by slowing just enough that a catastrophic packet loss never occurs … therefore sending data at the maximum possible speed for the maximum amount of time.”
“Since FastTCP uses only information already available in the TCP specification, it is 100 percent compatible with existing Internet infrastructure, and only needs installation on the sending end — intermediate hardware and software need not be modified,” O’Connell concluded. “Furthermore, as FastTCP does not cause the connection to cycle between too fast and too slow, it actually helps intermediate routers perform more efficiently, by not overwhelming them until the connection collapses and then slowing the connection.”