This paper presents the design philosophy and initial design decisions of Herald: a highly scalable global event notification system that is being designed and built at Microsoft Research. Herald is a distributed system designed to transparently scale in all respects, including numbers of subscribers and publishers, numbers of event subscription points, and event delivery rates. Event delivery can occur within a single machine, within a local network or Intranet, and throughout the Internet. Herald tries to take into account the lessons learned from the successes of both the Internet and the Web. Most notably, Herald is being designed, like the Internet, to operate correctly in the presence of numerous broken and disconnected components. The Herald service will be constructed as a set of protocols governing a federation of machines within cooperating but mutually suspicious domains of trust. Like the Web, Herald will try to avoid, to the extent possible, the maintenance of globally consistent state and will make failures part of the client-visible interface.

A network subsystem supporting a continuous media file system must guarantee a minimum throughput, a maximum delay, and a maximum jitter. The authors present a transport protocol that provides these guarantees. To support different types of service, the protocol is built from modules selected to meet the requirements of each communication session. A buffering technique is used to provide jitter guarantees. To provide throughput and delay guarantees, network performance is optimized based on the required transfer rate. The effects of controlling transmission rate and packet size are presented. The resulting transport protocol is modeled on a simulated FDDI (fiber distributed data interface) network and the results are analyzed. It is shown that the protocol provides the required guarantees for the anticipated types of traffic.>

This paper presents our experience with implementing atomicity in two systems: the QuickSilver distributed file system and the Starburst relational database manager. Each of these systems guarantees that certain collections of operations done on behalf of their clients execute atomically, despite process, machine, or network failures. In this paper we describe the atomic properties implemented by each system, present the algorithms and mechanisms used, examine the similarities and differences between the two systems, and give the rationale for different design decisions. We demonstrate that the support of atomicity with high performance requires a variety of techniques carefully chosen to balance the amount of data logged, the level of concurrency allowed, and the mutual consistency requirements of sets of objects, The main goal is to help others implement efficient systems that support atomicity.>

CACL is a protection scheme for objects that offers a simple and flexible model of protection and has an efficient, software-only implementation.The model, based on Access Control Lists (ACLs) integrated with the type system, allows owners to control who may invoke which methods on which objects, permits cooperation between mutually suspicious principals, allows ownership of objects to be transferred safely, prevents unwanted propagation of authority between principals, and allows changes to the authorization information to take effect on the next method invocation.The implementation, based on the integration of Capabilities with method dispatch, avoids the overhead of access checking in the majority of invocations, at the cost of space for extra dispatch vectors.CACL offers a viable mechanism for finegrained protection in an object-oriented database system.

As part of the Starburst extensible database project, the authors have designed and implemented a memory resident storage component that can coexist along side traditional disk-oriented storage components. The memory resident storage component shares the code of Starburst's common services, such as query optimization, plan generation, query evaluation, record manipulation, and transaction management. The design of Starburst's memory resident storage component is discussed, contrasted with Starburst's default disk-oriented storage component, and compared to the performance of the two storage components using the Wisconsin Benchmarks. The results show that a memory resident storage component can perform significantly better than a disk-oriented storage component, even when the disk-oriented storage component has all of its data cached in memory. The benchmark results show that, by using memory resident techniques, overall query execution can be improved by up to a factor of four.>

QuickSilver is a distributed operating system that is intended for large, geographically dispersed systems. The authors describe how clients and data servers interact in the QuickSilver system to access distributed data in the context of large, geographically dispersed systems in which there are many different kinds of data servers. An interface is defined that provides a uniform way for clients and servers to exchange information about each other and to access data objects. The interface attempts to minimize the number of messages needed for this. It supports efficient interaction with higher-level client and server semantics by appropriate use of high-level abstractions and operations, and by selectively exposing the parameters governing communication connections to higher layers of the software.>

The authors present a study of the performance achieved by user processes when using the IPC mechanisms as implemented in Berkeley Unix 4.2BSD in Ethernet based environments. The authors assess not only the impact that different processors, network hardware interfaces, and Ethernets have on the communication across machines, but also the effect of the loading of the hosts and communication media that participate in the interprocess communication mechanism. The measurements highlight the ultimate bounds on performance that may be achieved by user process applications communicating across machines, and serve as a guide in designing performance-critical applications. A detailed timing analysis is presented of the dynamic behavior of the TCP/IP and the UDP/IP network communication protocols' implementation in Berkeley Unix 4.2BSD.>