fliu@uvicctr.UVic.CA.UUCP (Frances Liu) (04/26/89)
I am working on a fault-tolerant remote execution manager system on network of SUN workstations. This system allows users to replicate processes on different nodes and uses pipe and socket for communication among processes. The communication is pair-wise between processes and is rather inefficient when a parent process needs to communicate with several replicate child processes simultaneously. We would like to implement a multicast facility in order to speedup communication among workstations. However the problem is the fact that the remote execution manager system is built entirely in the application layer. We are not supposed to change the unix kernel in the process of implementing this multicast facility. Is there anyone out there who has any ideas how it can be done? I am also interested in the algorithms and protocols that ensure reliability and atomicity of broadcast messages. Any ideas, comments and suggestions which relevant to the above issues are welcome. Thank you very much. Frances -- Frances Liu UUCP: ...!{ubc-vision,uw-beaver}!uvicctr!fliu APRAnet: fliu@uvicctr.nrl-css.arpa CDNnet: fliu@uvunix.uvic.cdn Internet: fliu%uvunix.uvic.ca@relay.ubc.ca BITNET: fliu@uvunix.bitnet