Shun Yan Cheung

On the use of destination set grouping to improve fairness in multicast video distribution

In a fair multicast video distribution scheme each receiver should receive a video stream with a quality that is commensurate with its capabilities or the capabilities of the path leading to it, regardless of other receivers or network paths. This fairness problem results from the fact that multicast communication trades economy of bandwidth with granularity of control. Distributing video using individual feedback-controlled point-to-point streams results in high bandwidth utilization but the granularity of control is high as communication parameters can be negotiated individually with each receiver. In contrast, using a single multicast stream has good bandwidth economy, but very low granularity of control. In this paper we propose, implement and experiment with a system that spans the spectrum represented by the two extremes above. In the scheme, called destination set grouping (DSG), a source maintains a small number of video streams, carrying the same video but each targeted at receivers with different capabilities. Each stream is feedback-controlled within prescribed limits by its group of receivers. Receivers may move among streams as their capabilities or the capabilities of the network paths leading to them change. The scheme is shown to improve fairness significantly at a small bandwidth cost.

The grid protocol: a high performance scheme for maintaining replicated data

A protocol for maintaining replicated data that can provide both high data availability and low response time is presented. Existing protocols are designed primarily to achieve high availability by updating a large fraction of the copies, which provides some (although not significant) load sharing. In the new protocol, transaction processing is shared effectively among nodes storing copies of the data, and both the response time experienced by transactions and the system throughput are improved significantly. Also presented is an analysis of the availability of the new protocol and simulation is used to study the effect of load sharing on the response time of transactions. The new protocol is also compared with a voting-based scheme.<<ETX>>

Routing multipoint connections using virtual paths in an ATM network

The point-to-multipoint routing problem is studied for an asynchronous transfer mode (ATM) network that uses virtual paths (VPs). ATM networks with asymmetric and symmetric VPs are considered, and the performance factors studied are bandwidth and establishment and switching costs. A VP with intermediate exit, where a node that performs VP switching can copy the switched packets for the local destination, is proposed and studied. Mathematical formulations of multicast routing problems are presented, and heuristics for finding a low cost multicast routing tree, based on the transshipment simplex algorithm, are developed.<<ETX>>

Multidimensional voting

We introduce a new concept, m ultidimenszonal voting, in which the vote and quorum assignments are k-dimensional vectors of nonnegative integers and each dimension is independent of the others. Multidimensional voting is more powerful than traditional weighted voting because it is equivalent to the general method for achieving synchronization in distributed systems which is based on sets of groups of nodes (quorum sets). We describe an efficient algorithm for finding a multidimensional vote assignment for any given quorum set and show examples of its use. We demonstrate the versatility of multidimensional voting by using it to implement mutual exclusion in fault-tolerant distributed systems and protocols for synchronizing access to fully and partially replicated data. These protocols cannot be implemented by traditional weighted voting. Also, the protocols based on multidimensional voting are easier to implement and/or provide greater flexibility than existing protocols for the same purpose, Finally, we present a generalization of the multidimensional voting scheme, called nested multidzmenszonal uotmg, that can facilitate implementation of replica control protocols that use structured quorum sets.

Using destination set grouping to improve the performance of window-controlled multipoint connections

In conventional multicast communication, the source carries a single conversation with all destination nodes. If a node on the path to any destination becomes congested, the throughput to all destinations is reduced, thus treating some destination nodes unfairly. We consider a window-controlled multipoint connection and study the use of destination set grouping, where the destination set can be split into disjoint subgroups with the source carrying independent conversations with each subgroup. We present a static grouping heuristic that can obtain near optimum grouping for static network environments and a dynamic grouping protocol which can adjust the grouping and the window sizes per group in response to changing network conditions. The performance of the static grouping heuristic and the dynamic grouping protocol are studied using simulation and compared with single-group multicasting.

BSFQ: bin sort fair queueing

Existing packet schedulers that provide fair sharing of an output link can be divided into two classes: sorted priority and frame-based. Sorted priority methods provide excellent approximation for weighted fair queueing (WFQ) while frame-based methods are more computationally efficient. We present a new packet scheduling algorithm called bin sort fair queueing (BSFQ) that combines the strengths of both type of schedulers. As a result, BSFQ is highly scalable and can provide very good approximation for WFQ. We prove that BSFQ can provide end-to-end delay and fairness guarantees to conformant flows. BSFQ also has a built-in buffer management function that can protect packets of conformant flows from nonconformant traffic. The performance of BSFQ and its ability to detect nonconformant flows are studied using simulations and compared to those of the deficit round robin method.

Multidimensional voting: a general method for implementing synchronization in distributed systems

A concept called multidimensional voting, in which the vote and quorum assignments are k-dimensional vectors of nonnegative integers and each dimension is independent of the others, is introduced. Multidimensional voting is more powerful than traditional weighted voting because it is equivalent to the general method for achieving synchronization in distributed systems which is based on coteries (sets of groups of nodes), but its implementation is easier than that of coteries. An efficient algorithm for finding a multidimensional vote assignment for any given coterie is described and examples of its use are shown. It is shown how multidimensional voting can be used to easily implement novel algorithms for synchronizing access to replicated data or to ensure mutual exclusion. These algorithms cannot be implemented by traditional weighted voting.<<ETX>>

Optimizing vote and quorum assignments for reading and writing replicated data

The problem is discussed of determining the vote assignment and quorum that yields the highest availability in a system where node availabilities can be different and the mix of the read and write operations is arbitrary. For this purpose, an enumeration algorithm is presented that can be used to find the vote and quorum assignments that need to be considered for achieving optimal availability. An analytical method is derived to evaluate the availability of a given system for any vote and quorum assignment. This method and the enumeration algorithm are used to find the optimal vote and quorum assignment for several systems. The algorithm can also be used to obtain the optimal performance when other measures are considered.<<ETX>>

Pipelined sections: a new buffer management discipline for scalable QoS provision

The techniques used to provide quality of service in packet switched networks are buffer management and packet scheduling. The first line of defense against abusive flows that transmit excessive number of packets is the buffer manager. When the number of packets of a flow exceeds a threshold, new arrivals from that flow will be rejected. This threshold must be set to a level that is sufficiently high to provide specific guarantees to the flow and it must also be minimal so that the buffer manager will detect abusive behavior at the earliest possible moment. A buffer management technique that provides the same guarantees using a lower threshold will be more discriminating because it recognizes flow violations earlier. We present the new pipelined sections buffer management technique that is highly scalable and can provide rate guarantees to leaky bucket constrained flows using very low buffer reservations.

Controlled request DQDB: achieving fairness and maximum throughput in the DQDB network

The author presents a new technique, called controlled request, to combat the unfairness problem in the distributed queue dual bus (DQDB) protocol that achieves 100% throughput. Transmission of requests in the controlled request DQDB method is controlled by a mechanism that is similar to the one used in DQDB to transmit segments. Fairness is achieved by using the bandwidth balancing technique to prevent an overloaded node form continuously transmitting requests. Simulation studies showed that the controlled request DQDB scheme achieves better performance than the DQDB protocol with bandwidth balancing.<<ETX>>