An Experiment in Latency Reduction

1,2 Communication latency can be reduced by increasing bandwidth via a sender-based anticipation technique called Parallel Communication. Here we apply this method to anonymous FTP. Our analysis of log files indicates that latency can be reduced to 2 round-trip times, as small as 0.6 round-trip time/file, for a 7x increase in bandwidth. This technique applies to up to 95% of the FTP traffic. This method is expected to be especially useful in reducing latency in automated FTP access, such as in the World-WideWeb. Current gigabit networking research addresses limitations of communication and computation " imposed by factors other than the finite speed of light " [10]. Communication limits computation via latency, and computation limits communication, especially regarding coding and compression [13]. There is a need for an " information theoretic theory of networks " [10]. Just as computation is analyzed at various levels of abstraction, from the Boolean/gate logic levels, through the abstract machine to algorithms, communication must go beyond the bit-level coding and communication theory, to an algo-rithmic abstraction. To this end, we are addressing the 'finite' limits of communication latency. Whereas the speed of light remains a fundamental limit, propagation latency alone is 1. not a limit to interaction latency in many cases. There are often ways to reduce the user-perceived communication latency to below that of speed-of-light propagation [13]. Network rate increases do not address the problem of propagation latency given effectively infinite bandwidth 3. What will this available bandwidth be used for? What will be changed as a result? We propose using excess band-width to reduce propagation latency. This assumes that bandwidth is a plentiful resource and latency is not. We are interested in considering bandwidth as a means, rather than an end. We want to investigate the real fundamental limit of communication-that of propagation latency, which is not currently being addressed [10]. From its inception in 1969 through 1988, available communication bandwidth remained constant in the Inter-net backbone as other computer resources grew. Band-width is resuming its own technology curve [13]. We notice that its rate of increase is challenging that of other computational resources, such as CPU rate, workstation memory, disk capacity, etc. On a per-year basis, workstation sizes increase at a rate of 1.26x [12]. DRAMs increase at a well-established rate of 1.59x, equivalent to 4x in 3 years. Disk capacity increases at 1.26x, equivalent to 2x in 3 years [5]. Microprocessors power …