Roy Friedman

Starfish: Fault-Tolerant Dynamic MPI Programs on Clusters of Workstations

This paper reports on the architecture and design of Starfish, an environment for executing dynamic (and static) MPI-2 programs on a cluster of workstations. Starfish is unique in being efficient, fault-tolerant, highly available, and dynamic as a system internally, and in supporting fault-tolerance and dynamicity for its application programs as well. Starfish achieves these goals by combining group communication technology with checkpoint/restart, and uses a novel architecture that is both flexible and portable and keeps group communication outside the critical data path, for maximum performance.

Starfish: fault-tolerant dynamic MPI programs on clusters of workstations

This paper reports on the architecture and design of Starfish, an environment for executing dynamic (and static) MPI-2 programs on a cluster of workstations. Starfish is unique in being efficient, fault-tolerant, highly available, and dynamic as a system internally, and in supporting fault-tolerance and dynamicity for its application programs as well. Starfish achieves these goals by combining group communication technology with checkpoint/restart, and uses a novel architecture that is both flexible and portable and keeps group communication outside the critical data path, for maximum performance.

A framework for protocol composition in Horus

The Horus system supports a communication architecture that treats protocols as instances of an abstract data type. This approach encourages developers to partition complex protocols into simple microprotocols, each of which is implemented by a protocol layer. Protocol layers can be stacked on top of each other in a variety of ways, at run-time. First, we describe the classes of protocols that can be supported this way. Next, we present the Horus object model that we designed for this technology, and the interface between the layers that makes it all work. We then present an example layer that implements a group membership protocol. Next, we show how, given a set of required properties, an appropriate stack can be constructed. We look at an example stack of protocols, which provides fault-tolerant, totally ordered communication between a group of processes. The work contributes a standard framework for protocol development and experimentation, provides a high performance implementation of the virtual synchrony model, and introduces a methodology for increasing the robustness of the protocol development process.

Strong and weak virtual synchrony in Horus

This paper presents two variants of virtual synchrony, which are supported by Horus. The first variant, called strong virtual synchrony, includes the property that every message is delivered within the view in which it is sent. This property is very useful in developing applications, since it helps in minimizing the amount of context information that needs to be sent on messages, and the amount of computation which is required in order to process a message. However, it is shown that in order to support this property, the application program has to block messages during view changes. An alternative definition, called weak virtual synchrony, which can be implemented without blocking messages, is then presented. This definition still guarantees that messages will be delivered within the view in which they were sent, only that it uses a slightly weaker notion of what the view in which a message was sent is. An implementation of weak virtual synchrony that does not block messages during view changes as also developed in this paper.

On Power and Throughput Tradeoffs of WiFi and Bluetooth in Smartphones

This paper describes a combined power and throughput performance study of WiFi and Bluetooth usage in smartphones. The work measures the obtained throughput in various settings while employing each of these technologies, and the power consumption level associated with them. In addition, the power requirements of Bluetooth and WiFi in their respective noncommunicating modes are also compared. The study reveals several interesting phenomena and tradeoffs. In particular, the paper identifies many situations in which WiFi is superior to Bluetooth, countering previous reports. The study also identifies a couple of scenarios that are better handled by Bluetooth. The conclusions from this study suggest preferred usage patterns, as well as operative suggestions for researchers and smartphone developers. This includes a cross-layer optimization for TCP/IP that could greatly improve the throughput to power ratio whenever the transmitter is more capable than the receiver.

RaWMS -: random walk based lightweight membership service for wireless ad hoc network

RaWMS is a novel lightweight random membership service for ad hoc networks. The service provides each node with a partial uniformly chosen view of network nodes. Such a membership service is useful, e.g., in data dissemination algorithms, lookup and discovery services, peer sampling services, and complete member-ship construction. The design of RaWMS is based on a novel re-verse random walk (RW) sampling technique. The paper includes a formal analysis of both the reverse RW sampling technique and RaWMS and verifies it through a detailed simulation study. In addition, RaWMS is compared with a number of other known methods such as flooding and gossip-based techniques.

Packing messages as a tool for boosting the performance of total ordering protocols

This paper compares the throughput and latency of four protocols that provide total ordering. Two of these protocols are measured with and without message packing. We used a technique that buffers application messages for a short period of time before sending them, so more messages are packed together. The main conclusion of this comparison is that message packing influences the performance of total ordering protocols under high load overwhelmingly more than any other optimization that was checked in this paper, both in terms of throughput and latency. This improved performance is attributed to the fact that packing messages reduces the header overhead for messages, the contention on the network, and the load on the receiving CPUs.

On power and throughput tradeoffs of WiFi and Bluetooth in smartphones

This paper describes a combined power and throughput performance study of WiFi and Bluetooth usage in smartphones. The work measures the obtained throughput in various settings while employing each of these technologies, and the power consumption level associated with them. In addition, the power requirements of Bluetooth and WiFi in their respective noncommunicating modes are also compared. The study reveals several interesting phenomena and tradeoffs. In particular, the paper identifies many situations in which WiFi is superior to Bluetooth, countering previous reports. The study also identifies a couple of scenarios that are better handled by Bluetooth. The conclusions from this study suggest preferred usage patterns, as well as operative suggestions for researchers and smartphone developers. This includes a cross-layer optimization for TCP/IP that could greatly improve the throughput to power ratio whenever the transmitter is more capable than the receiver.

RAPID: Reliable Probabilistic Dissemination in Wireless Ad-Hoc Networks

In this paper, we propose a novel reliable probabilistic dissemination protocol, RAPID, for mobile wireless ad-hoc networks that tolerates message omissions, node crashes, and selfish behavior. The protocol employs a combination of probabilistic forwarding with deterministic corrective measures. The forwarding probability is set based on the observed number of nodes in each one-hop neighborhood, while the deterministic corrective measures include deterministic gossiping as well as timer based corrections of the probabilistic process. These aspects of the protocol are motivated by a theoretical analysis that is also presented in the paper, which explains why this unique protocol design is inherent to ad-hoc networks environments. Since the protocol only relies on local computations and probability, it is highly resilient to mobility and failures. The paper includes a detailed performance evaluation by simulation. We compare the performance and the overhead of RAPID with the performance of other probabilistic approaches. Our results show that RAPID achieves a significantly higher node coverage with a smaller overhead.

Gossiping on MANETs: the beauty and the beast

Gossip protocols have emerged as a powerful technique for implementing highly scalable and robust services, such as information dissemination and aggregation. The fact that gossip protocols require very little or no structure to operate makes them particularly appealing to apply in dynamic systems, where topology changes are common (for instance, due to frequent faults or high churn rates). Therefore, gossip protocols seem particularly well fit to operate in wireless self-organizing networks. Unfortunately, these networks have a number of characteristics that impede the deployment of gossip protocols designed for wired networks. In this work we identify the inherent differences in communication between wired and wireless networks and their impact on the design and implementation of gossip protocols. In particular, our comparison includes drawing a distinction between the gossiping primitives suitable for each of these environments. In the context of this analysis, we conclude by presenting a list of open research questions.