NCS: Message Transfer Systems
|Prev: Alice Intro||Course Home||Next: Pthreads|
This lecture describes different choices for message transfer systems in a networked control system. The message transfer system is responsible for managing network communications between computers and software modules with the control system. We focus on systems that are build on top of the TCP/IP protocol stack. Design choices include how to encode information in packets, whether to broadcast or send packets point-to-point, and whether to retransmit packets on lost data transmission. Because of the closed loop nature of the networked embedded systems that we are programming, timing and latency are critical issues. We focus on the use of spread as a specific example of a low-level message transfer subsystem and describe how it can be used in a NCS context.
Time, clocks, and the ordering of events in a distributed system, L. Lamport. Communications of the ACM, 21(7):558-565, 1978. This is a classic paper on ordering of messages in distributed systems. A must read for distributed systems.
Exploiting virtual synchrony in distributed systems, K. Birman and T. Joseph. ACM Symposium on Operating Systems Principles, 1987. This paper gives a nice overview of some of the problems in group messaging systems and is one of the sets of papers that motivated the work that lead to Spread.
A Users Guide to Spread, J. Stanton. 2002. This is the documentation for the Spread Toolkit. The first and second chapters provide most of the information you need to understand the basic ideas, although the way in which Spread servers are configured, described in Chapter 3, is also useful.
The Spread Wide Area Group Communication System, Y. Amir and J. Stanton. Technical Report CNDS-98-4, The Center for Networking and Distributed Systems, The Johns Hopkins University, 1998. The paper provides a detailed technical description of how Spread works. It is mainly useful if you want to know more about what spread does. Requires some background in network protocols if you want to understand the details.
A Low Latency, Loss Tolerant Architecture and Protocol for Wide Area Group Communication, Yair Amir, Claudiu Danilov, and Jonathan Stanton. International Conference on Dependable Systems and Networks (DSN00), New York, New York, June 25-28, 2000. The paper is similar to the one above but with more technical details.
- Spread Toolbox - The Spread Wide Area Group Communication System
- Neutral Message Language (NML) Programmers Guide - part of the NIST Realtime Control System library
- RoboFlag documentation - includes discussion of how RoboFlag network communications are implemented
- Modify the Spread protocol such that it can deal with the following situation. For typical control applications, most recent data is more important than the old data. Suppose that a process needs three packets, m1, m2 and m3, where m1 is older than m2 and m2 is older than m3. Suppose that m2 is missing, then the process should only request for this packet on the condition that m3 is not received yet. Current version of Spread has not addressed this issue yet.