Jones, Mark G. W.; (1993) The Scalability of Multicast Communication. Masters thesis (M.Phil), UCL (University College London).

Preview

Text The_scalability_of_multicast_c.pdf Download (5MB) | Preview

Abstract

Multicast is a communication method which operates on groups of applications. Having multiple instances of an application which are addressed collectively using a unique, multicast address, allows elegant solutions to some of the more intractable problems in distributed programming, such as providing fault tolerance. However, as multicast techniques are applied in areas such as distributed operating systems, where the operating system may span a large number of hosts, or on faster network architectures, where the problems of congestion reduce the effectiveness of the technique, then the scalability of multicast must be addressed if multicast is to gain a wider application. The main scalability issue was considered to be packet loss due to buffer overrun, the most common cause of this buffer overrun being the mismatch in packet arrival rate and packet consumption at the multicast originator, the so-called implosion problem. This issue affects positively acknowledged and transactional protocols. As these two techniques are the most common protocol designs, it was felt that an investigation into the problems of these types of protocol would be most effective. A model for implosion was developed which was simulated in order to investigate the parameters of implosion. A measure of this implosion was derived from the data, this index of implosion allowing the severity of implosion to be described as well as the location of the implosion in the model. This implosion index was derived by dividing the rate at which buffers were occupied by the rate at which packets were generated by the model. The value may then be used to predict the number of buffers required given the number of packets expected. A number of techniques were developed which may be used to offset implosion, either by artificially increasing the inter-packet gap, or by distributing replies so that no one host receives enough packets to cause an implosion. Of these alternatives, the latter offers the most promise, although requiring a large effort to maintain the resulting hierarchical structure in the presence of multiple failures.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item