Dalia Malki

The Transis approach to high availability cluster communication

In the local elections system of the municipality of “Wiredville”,1 several computers were used to establish an electronic town hall. The computers were linked by a network. When an issue was put to a vote, voters could manually feed their votes into any of the computers, which replicated the updates to all of the other computers. Whenever the current tally was desired, any computer could be used to supply an up-to-the-moment count. On the night of an important election, a room with one of the computers became crowded with lobbyists and politicians. Unexpectedly, someone accidentally stepped on the network wire, cutting communication between two parts of the network. The vote counting stopped until the network was repaired, and the entire tally had to be restarted from scratch. This would not have happened if the vote-counting system had been built with partitions in mind. After the unexpected severance, vote counting could have continued at all the computers, and merged appropriately when the network was repaired. The “Wiredville” story illustrates some of the finer points that motivated our work in the Transis project [1], a large-scale multicast service designed with the following goals:

Transis: a communication subsystem for high availability

The authors describe Transis, a communication subsystem for high availability. Transis is a transport layer that supports reliable multicast services. The main novelty is in the efficient implementation using broadcast. The basis of Transis is automatic maintenance of dynamic membership. The membership algorithm is symmetrical, operates within the regular flow of messages, and overcomes partitions and remerging. The higher layer provides various multicast services for sets of processes.<<ETX>>

Membership Algorithms for Multicast Communication Groups

We introduce a membership protocol that maintains the set of currently connected machines in an asynchronous and dynamic environment. The protocol handles both failures and joining of machines. It operates within a multicast communication sub-system.

A high-throughput secure reliable multicast protocol

A reliable multicast protocol enables a process to multicast a message to a group of processes in a way that ensures that all honest destination-group members receive the same message, even if some group members and the multicast initiator are maliciously faulty. Reliable multicast has been shown to be useful for building multiparty cryptographic protocols and secure distributed services. We present a high-throughput reliable multicast protocol that tolerates the malicious behavior of up to fewer than one-third of the group members. Our protocol achieves high-throughput using a novel technique for chaining multicasts, whereby the cost of ensuring agreement on each multicast message is amortized over many multicasts. This is coupled with a novel flow-control mechanism that yields low multicast latency.

A framework for partitionable membership service

This paper presents a framework for a membership service that operates in a partitionable environment and supports partitionable operation, which is a form of distributed operation in which multiple network components that are (temporarily) disconnected from each other operate autonomously. The service assumes an asynchronous environment and must tolerate crash failures, omission failures and network partitions. The principles of partitionable operation that we present here have been incorporated in the Transis system [13, 1], the Totem system [3], and the Horus system [19]. The paper discusses applications built in these projects, and relates them to the membership service de nition. We introduce a distinction between partial and complete installations of system views that makes feasible what we believe are the strongest possible requirements for causal order and virtual synchrony. We propose our speci cation of partitionable membership service as a standard against which other membership services can be measured.

Using visualization tools to understand concurrency

A visualization tool that provides an aggregate view of execution through a graph of events called the causality graph, which is suitable for systems with hundreds or thousands of processors, coarse-grained parallelism, and for a language that makes communication and synchronization explicit, is discussed. The methods for computing causality graphs and stepping through an execution with causality graphs are described. The properties of the abstraction algorithms and super nodes, the subgraphs in causality graphs, are also discussed.<<ETX>>

The Design of the Transis System

Transis is a high availability distributed system, being developed in the Hebrew University. It supports reliable group communication for high availability applications. The system provides enhanced services for information dissemination and replication in a dynamic environment where machines may crash, for arbitrarily long periods, and may recover; where the network may partition and re-merge. Transis contains novel protocols for reliable message delivery, it optimizes the performance for existing network hardware, and offers a variety of different handles to upper applications. The paper presents the experience gained in the design and the implementation of the Transis communication subsystem.

On Distributed Algorithms in a Broadcast Domain

This paper studies the usage of broadcast communication in distributed services. The approach taken is practical: all the algorithms are asynchronous, and tolerate realistic faults. We study four problems in a broadcast domain: clock synchronization, reliable and ordered broadcast, membership, and file replication. The clock synchronization algorithm shows for the first time how to utilize broadcast communication for synchronization. The master synchronizes any number of slaves while incurring a constant load. The approach taken in the file replication tool uses snooping in order to enhance the availability of file systems, at almost no cost.

Warm backup using snooping

Local area networks use a broadcast media to transfer messages between hosts. This allows for network snooping by unlisted parties. This paper proposes a novel way for cheaply replicating services in a local area network via snooping. We present a tool for the warm backup of files that employs network snooping for data dissemination. The tool allows for a selective replication of files in the system. The use of snooping significantly reduces the overhead of file replication. Operations on non-replicated files suffer only a slight overhead.<<ETX>>

Nicke-C extensions for programming on distributed-memory machines

Abstract This document describes the Nicke programming language. Nicke is an extension of C for programming distributed-memory machines. It supports both message-passing and shared-memory parallelism. We present the rationale for Nicke and describe the main features of the language. The Nicke language has been implemented on the 8CE machine under the Mach Operating System, [6], and is being ported to the Victor machine under the Trollius kernel [11, 5]. The language is supported by a preprocessor that translates Nicke code into standard C, and by a run-time library.