Scalable Stability Detection Using Logical Hypercube
IEEE Transactions on Parallel and Distributed Systems
Scalable Stability Detection Using Logical Hypercube
SRDS '99 Proceedings of the 18th IEEE Symposium on Reliable Distributed Systems
| Hi-index | 0.03 |
Protocols which track message stability are an important part of reliable multicast protocols in fault-tolerant distributed systems. To reliably deliver multicast messages in a process group, each process maintains copies of all messages it sends and receives. If a member fails to receive a message, any process which has the message in its buffer can retransmit it. In order to prevent these buffers from growing out of bound, stability tracking protocols must be used. That is, whenever a process learns that a message has been received by everyone, it declares this message {\it stable} and releases it from the buffer. We investigate several message stability tracking protocols commonly used in a number of popular reliable multicast protocols with a focus on their performance in large scale settings with thousands of participants. To improve the scalability of these protocols significantly, we derive a set of new protocols using a spanning tree structure which scale to at least tens of thousands of participants.