Fujitsu Laboratories claims to have developed a new software-only version of the Transmission Control Protocol (TCP) that has already achieved a “throughput increase of over 30 times” during simulated file transfers between Japan and the USA and it also improves latency (i.e. better ping times).
The standard TCP protocol, which practically everybody uses alongside an Internet Protocol (IP) address, often without even realising, was designed for the internet to help get data from one network device to another. Sadly TCP is old and its ability to adapt when congestion strikes at choke points on a network is somewhat limited.
By comparison the new method, which we’ll call Fujitsu TCP (FTCP) because they haven’t actually told us what it’s called, appears to take a leaf out of TCP’s cousin – the User Datagram Protocol. UDP performs a similar task to TCP but it lacks some of the error checking and correction, which in some respects makes it faster but also doesn’t guarantee data delivery (e.g. fast-paced multiplayer games often use UDP because it’s less prone to high latency).
Explanation of the New FTCP
Fujitsu has developed a new protocol that incorporates a proprietarily developed and efficient retransmission method based on UDP, a protocol optimized for delivering streaming media. As a result, the new protocol is able to reduce latency resulting from data retransmission when packet loss occurs.
The protocol can quickly distinguish between lost packets and packets that have not yet arrived at their destination, thereby preventing unnecessary retransmissions and latency from occurring.
By incorporating the new protocol as a software add-on to UDP, it is possible to maintain the high speeds typical of UDP while avoiding packet loss and packets being sent in reverse order, UDP’s main weaknesses. This, in turn, has enabled improvements in packet delivery and latency.
Now, to answer the obvious question, does this mean that your broadband connection will suddenly be able to download internet files 30 times faster once it’s introduced? Probably not. Certainly the new protocol could help to iron out some of the problems and improve TCP efficiency but it’s perhaps misleading to say that it would deliver such a huge physical boost to actual connection speeds (a much smaller boost.. perhaps).
It should also be stressed that Fujitsu are by no means the first firm to try and improve TCP, with solutions like TCPCT, TCPFP and more recently MPTCP all working to better the old standard. One problem that many of these methods share is that not all of them have been fully adopted and without strong adoption you don’t get the full benefit. Suffice to say that there are lot of technical complications to add here but we don’t have time to write a new book.
Fujitsu hopes to commercialise the new technology during 2013 as a “communications middleware solution” for improving communications speeds “without having to modify existing TCP applications“. We look forward to seeing the results.