TCP-XM
Karl Jeacle
Ph.D, University of Cambridge
September 2005
Supervisor: Prof Jon Crowcroft
Abstract
Almost 20 years since its inception, IP multicast lags far behind IP unicast in network deployment. While large clouds of native IP multicast enabled networks exist, end-to-end coverage is limited.
Without a killer application to drive deployment, multicast finds itself in a Catch-22 situation. Applications don't take advantage of multicast because the network doesn't support it, while network support is lacking because the applications don't demand it.
Because of this, multicast-enabled clouds carry relatively small amounts of multicast in comparison to unicast. Group communication almost always takes place using unicast transmission.
There has been much work carried out in recent years on reliable multicast. But the focus has typically been on scaling to large numbers of receivers, and assumptions are made regarding the state of multicast coverage. Consequently, even for small numbers of receivers, such as those found in Grid environments, few practical solutions are available today.
I would like a practical solution suitable for small-scale reliable group communication while also overcoming, to some extent, the deployment issue by making the best possible use of any partial network coverage.
Depending on the state of unicast and multicast connectivity between source and destinations, the right combination of unicast and multicast transmission can provide a large increase in efficiency at the cost of a small decrease in speed. Even the introduction of partial multicast connectivity can provide benefits.
Applications can initiate fully reliable data transfers while transparently taking full advantage of any deployed multicast connectivity and thus lower network utilisation.
With a clear goal of reliable multicast data transfer to a moderate number of receivers, I propose a novel hybrid user-space solution combining unicast TCP and multicast transmission simultaneously. Multicast transmission is always attempted, but fallback to unicast takes place should multicast fail.
I have taken my work further than many other reliable multicast protocols as I have designed, implemented, deployed and evaluated my solution.
Download thesis: 1.2MB PDF file
Papers:
Hybrid Reliable Multicast with TCP-XM (pdf paper, ppt slides)
ACM CoNEXT 2005
TCP-XM: Unicast-enabled Reliable Multicast (pdf paper, ppt slides)
IEEE ICCCN 2005
mcp: A Reliable Multicast File Transfer Tool (pdf abstract, ps
poster)
All Hands Meeting 2005
A Multicast Transport Driver for Globus XIO (pdf paper, ppt slides)
IEEE WETICE ETNGRID 2005
Extending Globus to support Multicast Transmission (pdf paper, ppt
slides)
All Hands Meeting 2004
Reliable Multicast for the Grid (doc abstract, pdf poster)
PREP2004
Reliable high-speed Grid data delivery using IP multicast
(pdf paper, ppt slides, jpg
poster)
All Hands Meeting 2003
TCP-XM: Multicast Transport for Grid Computing (ppt slides)
Multi-Service Networks 2003
Code:
