Motivated by the explosive growth of the Internet, we study efficient and fault-tolerant distributed session layer \nprotocols for networking elements. These protocols are \ndesigned to enable a network cluster to share the state \ninformation necessary for balancing network traffic and \ncomputation load among a group of networking elements. \nIn addition, in the presence of failures, they allow \nnetwork traffic to fail-over from failed networking \nelements to healthy ones. To maximize the overall \nnetwork throughput of the networking cluster, we assume a unicast communication medium for these protocols. The Raincore Distributed Session Service is based on a fault-tolerant token protocol, and provides group membership, reliable multicast and mutual exclusion services in a networking environment. We show that this service provides atomic reliable multicast with consistent ordering. We also show that Raincore token protocol consumes less overhead than a broadcast-based protocol in this environment in terms of CPU task-switching. The Raincore technology was transferred to Rainfinity, a startup company that is focusing on software for Internet reliability and performance. Rainwall, Rainfinity’s first product, was developed using the Raincore Distributed Session Service. We present initial performance results of the Rainwall product that validates our design assumptions and goals.

The RAIN project is a research collaboration between Caltech and NASA-JPL on distributed computing and data-storage systems for future spaceborne missions. The goal of the project is to identify and develop key building blocks for reliable distributed systems built with inexpensive off-the-shelf components. The RAIN platform consists of a heterogeneous cluster of computing and/or storage nodes connected via multiple interfaces to networks configured in fault-tolerant topologies. The RAIN software components run in conjunction with operating system services and standard network protocols. Through software-implemented fault tolerance, the system tolerates multiple node, link, and switch failures, with no single point of failure. The RAIN-technology has been transferred to Rainfinity, a start-up company focusing on creating clustered solutions for improving the performance and availability of Internet data centers. In this paper, we describe the following contributions: 1) fault-tolerant interconnect topologies and communication protocols providing consistent error reporting of link failures, 2) fault management techniques based on group membership, and 3) data storage schemes based on computationally efficient error-control codes. We present several proof-of-concept applications: a highly-available video server, a highly-available Web server, and a distributed checkpointing system. Also, we describe a commercial product, Rainwall, built with the RAIN technology.

Adams and Siegel (1982) proposed an extra stage cube interconnection network that tolerates one switch failure with one extra stage. We extend their results and discover a class of extra stage interconnection networks that tolerate multiple switch failures with a minimal number of extra stages. Adopting the same fault model as Adams and Siegel, the faulty switches can be bypassed by a pair of demultiplexer/multiplexer combinations. It is easy to show that, to maintain point to point and broadcast connectivities, there must be at least S extra stages to tolerate I switch failures. We present the first known construction of an extra stage interconnection network that meets this lower-bound. This 12-dimensional multistage interconnection network has n+f stages and tolerates I switch failures. An n-bit label called mask is used for each stage that indicates the bit differences between the two inputs coming into a common switch. We designed the fault-tolerant construction such that it repeatedly uses the singleton basis of the n-dimensional vector space as the stage mask vectors. This construction is further generalized and we prove that an n-dimensional multistage interconnection network is optimally fault-tolerant if and only if the mask vectors of every n consecutive stages span the n-dimensional vector space.