Lightweight Causal and Atomic Group Multicast The ISIS toolkit supports distributed programming using virtual synchronous process groups and group communication. This paper presents new protocols, including CBCAST for fault-tolerant, causally ordered message delivery and ABCAST for totally ordered multicast. CBCAST delivers messages immediately and incurs minimal overhead, while ABCAST extends CBCAST to ensure total order. These protocols enable performance and scalability comparable to raw message transport layers, contradicting concerns about high cost. The ISIS system supports four types of process groups: peer, client/server, diffusion, and hierarchical. These groups are used in various applications, with membership changes infrequent and group sizes typically small. The system's performance is heavily influenced by CBCAST, which is used extensively in Isis tools. The paper discusses the design and implementation of CBCAST and ABCAST protocols. CBCAST uses vector timestamps to ensure causal delivery order, while ABCAST extends this to total order. The protocols are efficient, with ABCAST being more costly than CBCAST but still scalable. The protocols support concurrent applications, large numbers of overlapping groups, and bound overhead to group size. The paper also addresses issues such as multicast stability, vector time compression, and handling group membership changes. It introduces a flush protocol to ensure virtual synchrony during group membership changes, and discusses the atomicity of multicast delivery in the presence of failures. The protocols are implemented with a focus on efficiency and scalability, allowing the system to support large-scale applications. The new protocols are more efficient than earlier solutions, enabling higher performance and scalability. The paper concludes with a discussion of the performance of the initial implementation and the potential for future improvements.Lightweight Causal and Atomic Group Multicast The ISIS toolkit supports distributed programming using virtual synchronous process groups and group communication. This paper presents new protocols, including CBCAST for fault-tolerant, causally ordered message delivery and ABCAST for totally ordered multicast. CBCAST delivers messages immediately and incurs minimal overhead, while ABCAST extends CBCAST to ensure total order. These protocols enable performance and scalability comparable to raw message transport layers, contradicting concerns about high cost. The ISIS system supports four types of process groups: peer, client/server, diffusion, and hierarchical. These groups are used in various applications, with membership changes infrequent and group sizes typically small. The system's performance is heavily influenced by CBCAST, which is used extensively in Isis tools. The paper discusses the design and implementation of CBCAST and ABCAST protocols. CBCAST uses vector timestamps to ensure causal delivery order, while ABCAST extends this to total order. The protocols are efficient, with ABCAST being more costly than CBCAST but still scalable. The protocols support concurrent applications, large numbers of overlapping groups, and bound overhead to group size. The paper also addresses issues such as multicast stability, vector time compression, and handling group membership changes. It introduces a flush protocol to ensure virtual synchrony during group membership changes, and discusses the atomicity of multicast delivery in the presence of failures. The protocols are implemented with a focus on efficiency and scalability, allowing the system to support large-scale applications. The new protocols are more efficient than earlier solutions, enabling higher performance and scalability. The paper concludes with a discussion of the performance of the initial implementation and the potential for future improvements.