Tallat M. Shafaat

A Practical Approach to Network Size Estimation for Structured Overlays

Structured overlay networks have recently received much attention due to their self-* properties under dynamic and decentralized settings. The number of nodes in an overlay fluctuates all the time due to churn. Since knowledge of the size of the overlay is a core requirement for many systems, estimating the size in a decentralized manner is a challenge taken up by recent research activities. Gossip-based Aggregation has been shown to give accurate estimates for the network size, but previous work done is highly sensitive to node failures. In this paper, we present a gossip-based aggregation-style network size estimation algorithm. We discuss shortcomings of existing aggregation-based size estimation algorithms, and give a solution that is highly robust to node failures and is adaptive to network delays. We examine our solution in various scenarios to demonstrate its effectiveness.

Key-based consistency and availability in structured overlay networks

Structured Overlay Networks (SONs) provide a promising platform for high performance applications since they are scalable, fault-tolerant and self-managing. SONs provide lookup services that map keys to nodes that can be used as processing or storage resources. In SONs, lookups for a key may return inconsistent results. Consequently, it is difficult to provide consistent data services on top of SONs that build on key-based search. In this paper, we study the frequency of occurrence of inconsistent lookups. We show that the affect of lookup inconsistencies can be reduced by using node responsibilities. We present our results as a trade-off between consistency and availability of keys.

Dealing with network partitions in structured overlay networks

Structured overlay networks form a major class of peer-to-peer systems, which are touted for their abilities to scale, tolerate failures, and self-manage. Any long-lived Internet-scale distributed system is destined to face network partitions. Although the problem of network partitions and mergers is highly related to fault-tolerance and self-management in large-scale systems, it has hardly been studied in the context of structured peer-to-peer systems. These systems have mainly been studied under churn (frequent joins/failures), which as a side effect solves the problem of network partitions, as it is similar to massive node failures. Yet, the crucial aspect of network mergers has been ignored. In fact, it has been claimed that ring-based structured overlay networks, which constitute the majority of the structured overlays, are intrinsically ill-suited for merging rings. In this paper, we present an algorithm for merging multiple similar ring-based overlays when the underlying network merges. We examine the solution in dynamic conditions, showing how our solution is resilient to churn during the merger, something widely believed to be difficult or impossible. We evaluate the algorithm for various scenarios and show that even when falsely detecting a merger, the algorithm quickly terminates and does not clutter the network with many messages. The algorithm is flexible as the tradeoff between message complexity and time complexity can be adjusted by a parameter.

Handling Network Partitions and Mergers in Structured Overlay Networks

Structured overlay networks form a major class of peer- to-peer systems, which are touted for their abilities to scale, tolerate failures, and self-manage. Any long-lived Internet-scale distributed system is destined to face network partitions. Although the problem of network partitions and mergers is highly related to fault-tolerance and self-management in large-scale systems, it has hardly been studied in the context of structured peer-to-peer systems. These systems have mainly been studied under churn (frequent joins/failures), which as a side effect solves the problem of network partitions, as it is similar to massive node failures. Yet, the crucial aspect of network mergers has been ignored. In fact, it has been claimed that ring-based structured overlay networks, which constitute the majority of the structured overlays, are intrinsically ill-suited for merging rings. In this paper, we present an algorithm for merging multiple similar ring-based overlays when the underlying network merges. We examine the solution in dynamic conditions, showing how our solution is resilient to churn dur- ng the merger, something widely believed to be difficult or mpossible. We evaluate the algorithm for various scenar- os and show that even when falsely detecting a merger, the algorithm quickly terminates and does not clutter the network with many messages. The algorithm is flexible as the tradeoff between message complexity and time complexity can be adjusted by a parameter.

ID-replication for structured peer-to-peer systems

Structured overlay networks, like any distributed system, use replication to avoid losing data in the presence of failures. In this paper, we discuss the short-comings of existing replication schemes and propose a technique for replication, called ID-Replication. ID-Replication allows different replication degrees for keys in the system, thus allowing popular data to have more copies. We discuss how ID-Replication is less sensitive to churn compared to existing replication schemes, which makes ID-Replication better suited for building consistent services on top of overlays compared to other schemes. Furthermore, we show why ID-Replication is simpler to load-balance and more secure compared to successor-list replication. We evaluate our scheme in detail, and compare it with successor-list replication.

On Consistency Of Data In Structured Overlay Networks

Data consistency can be violated in Distributed Hash Tables (DHTs) due to inconsistent lookups. In this paper, we identify the events leading to inconsistent lookups and inconsistent responsibilities for a key. We find the inaccuracy of failure detectors as the main reason for inconsistencies. By simulations with inaccurate failure detectors, we study the probability of reaching a system configuration which may lead to inconsistent data. We analyze majority-based algorithms for operations on replicated data. To ensure that concurrent operations do not violate consistency, they have to use non-disjoint sets of replicas. We analytically derive the probability of concurrent operations including disjoint replica sets. By combining the simulation and analytical results, we show that the probability for a violation of data consistency is negligibly low for majority-based algorithms in DHTs.

A method of adaptive coarsening for compressing scientific datasets

We present adaptive coarsening, a multi-resolution lossy compression algorithm for scientific datasets. The algorithm provides guaranteed error bounds according to the users requirements for subsequent post-processing. We demonstrate compression factors of up to an order of magnitude with datasets coming from solutions to timedependent partial differential equations in one and two dimensions.

Brief announcement: atomic consistency and partition tolerance in scalable key-value stores

We propose consistent quorums to achieve linearizability in scalable and self-organizing key-value stores based on consistent hashing.

CATS: a linearizable and self-organizing key-value store

Distributed key-value stores provide scalable, fault-tolerant, and self-organizing storage services, but fall short of guaranteeing linearizable consistency in partially synchronous, lossy, partitionable, and dynamic networks, when data is distributed and replicated automatically by the principle of consistent hashing [14]. This work introduces consistent quorums as a solution for achieving atomic consistency. We present the design and implementation of CATS, a key-value store which uses consistent quorums to guarantee linearizability and partition tolerance in such adverse and dynamic network conditions. CATS is scalable, elastic, and self-organizing; key properties for modern cloud storage middleware. Our system evaluation shows that consistency can be achieved with practical performance and modest overhead: 5% decrease in throughput for read-intensive workloads, and 25% throughput loss for write-intensive workloads. CATS delivers submillisecond operation latencies under light load, single-digit millisecond operation latencies at 50% load, and it sustains a throughput of one thousand operations per second, per server, while scaling linearly to hundreds of servers.

Dealing with Bootstrapping, Maintenance, and Network Partitions and Mergers in Structured Overlay Networks

In the last decade, numerous structured overlay networks were proposed as a scalable infrastructure to build large-scale distributed systems under dynamic environments. These overlays were touted to be fault-tolerant and self-managing, yet, as we show in this paper, they fall short of handling some extreme scenarios they envision. These scenarios include bootstrapping, and underlying network partitions and mergers. We argue that handling such extreme scenarios is fundamental to providing a fault-tolerant and self-managing system, and thus, structured overlay networks should intrinsically be able to handle them. In this paper, we present ReCircle, an overlay algorithm that apart from performing periodic maintenance to handle churn like any other overlay, can merge multiple structured overlay networks. We show how such an algorithm can be used for decentralized bootstrapping. ReCircle does not have any extra cost during normal maintenance compared to an isolated overlay maintenance algorithm. Furthermore, the algorithm is tunable to tradeoff between bandwidth consumption and time to convergence during extreme events like bootstrapping and handling network partitions and mergers. We evaluate the algorithm extensively under various scenarios through simulation and experimentation on Planet Lab.