Boffins at Google are looking at ways of speeding up the internet. Not by making the underlying physical network faster, but by improving the efficiency of the Transmission Control Protocol (TCP) that underpins the internet. This will have the benefit of making all networks that use the new improved version faster and being more scalable with improvements to the underlying physical network. The new TCP stack will be compatible with the old one by running alongside it and only kicking in when both client and server have it installed. This is currently implemented in the Linux kernel and is in the process of becoming the TCP standard.

There are four basic ways that they're looking at improving TCP: 1) Increase the number of TCP packets sent at the beginning of a TCP connection from three to ten. This reduces latency. 2) Reduce packet loss timeout from three seconds to one ie shorten the time between retries, which is more appropriate for modern high speed networks. 3) Use TCP Fast Open (TFO) a faster way for a web browser to initiate a connection with a website and 4) Use Proportional Rate Reduction for TCP (PRR) which is a more efficient way to deal with traffic congestion.

The research team are also looking at improved transmission recovery (data loss) in noisy mobile environments. Full technical details are available at the Google blog.