We propose a family of constant-degree routing networks of logarithmic diameter, with the additional property that the addition or removal of a node to the network requires no global coordination, only a constant number of linkage changes in expectation, and a logarithmic number with high probability. Our randomized construction improves upon existing solutions, such as balanced search trees, by ensuring that the congestion of the network is always within a logarithmic factor of the optimum with high probability. Our construction derives from recent advances in the study of peer-to-peer lookup networks, where rapid changes require e#cient and distributed maintenance, and where the lookup e#ciency is impacted both by the lengths of paths to requested data and the presence or elimination of bottlenecks in the network.

Today an application developer using a distributed hash table (DHT) with n nodes must choose a DHT protocol from the spectrum between O(1) lookup protocols [9, 18] and O(log n) protocols [20–23,25,26]. O(1) protocols achieve low latency lookups on small or low-churn networks because lookups take only a few hops, but incur high maintenance traffic on large or high-churn networks. O(log n) protocols incur less maintenance traffic on large or highchurn networks but require more lookup hops in small networks. Accordion is a new routing protocol that does not force the developer to make this choice: Accordion adjusts itself to provide the best performance across a range of network sizes and churn rates while staying within a bounded bandwidth budget. The key challenges in the design of Accordion are the algorithms that choose the routing table’s size and content. Each Accordion node learns of new neighbors opportunistically, in a way that causes the density of its neighbors to be inversely proportional to their distance in ID space from the node. This distribution allows Accordion to vary the table size along a continuum while still guaranteeing at most O(log n) lookup hops. The user-specified bandwidth budget controls the rate at which a node learns about new neighbors. Each node limits its routing table size by evicting neighbors that it judges likely to have failed. High churn (i.e., short node lifetimes) leads to a high eviction rate. The equilibrium between the learning and eviction processes determines the table size. Simulations show that Accordion maintains an efficient lookup latency versus bandwidth tradeoff over a wider range of operating conditions than existing DHTs.

Typical algorithms for decentralized data distribution work best in a system that is fully built before it first used; adding or removing components results in either extensive reorganization of data or load imbalance in the system. We have developed a family of decentralized algorithms, RUSH (Replication Under Scalable Hashing), that maps replicated objects to a scalable collection of storage servers or disks. RUSH algorithms distribute objects to servers according to user-specified server weighting. While all RUSH variants support addition of servers to the system, different variants have different characteristics with respect to lookup time in petabyte-scale systems, performance with mirroring (as opposed to redundancy codes), and storage server removal. All RUSH variants redistribute as few objects as possible when new servers are added or existing servers are removed, and all variants guarantee that no two replicas of a particular object are ever placed on the same server. Because there is no central directory, clients can compute data locations in parallel, allowing thousands of clients to access objects on thousands of servers simultaneously.

As storage systems scale to thousands of disks, data distribution and load balancing become increasingly important. We present an algorithm for allocating data objects to disks as a system as it grows from a few disks to hundreds or thousands. A client using our algorithm can locate a data object in microseconds without consulting a central server or maintaining a full mapping of objects or buckets to disks. Despite requiring little global configuration data, our algorithm is probabilistically optimal in both distributing data evenly and minimizing data movement when new storage is added to the system. Moreover, our algorithm supports weighted allocation and variable levels of object replication, both of which are needed to permit systems to efficiently grow while accommodating new technology. 1

In many decision making applications, users typically issue aggregate queries. To evaluate these computationally expensive queries, online aggregation has been developed to provide approximate answers (with their respective confidence intervals) quickly, and to continuously refine the answers. In this paper, we extend the online aggregation technique to a distributed context where sites are maintained in a DHT (Distributed Hash Table) network. Our Distributed Online Aggregation (DoA) scheme iteratively and progressively produces approximate aggregate answers as follows: in each iteration, a small set of random samples are retrieved from the data sites and distributed to the processing sites; at each processing site, a local aggregate is computed based on the allocated samples; at a coordinator site, these local aggregates are combined into a global aggregate. DoA adaptively grows the number of processing nodes as the sample size increases. To further reduce the sampling overhead, the samples are retained as a precomputed synopsis over the network to be used for processing future queries. We also study how these synopsis can be maintained incrementally. We have conducted extensive experiments on PlanetLab. The results show that our DoA scheme reduces the initial waiting time significantly and provides high quality approximate answers with running confidence intervals progressively. 1.

Abstract not found

Abstract — This paper presents a new parallel indexing data structure for answering queries. The index, called Bin-Hash, offers extremely high levels of concurrency, and is therefore wellsuited for the emerging commodity of parallel processors, such as multi-cores, cell processors, and general purpose graphics processing units (GPU). The Bin-Hash approach first bins the base data, and then partitions and separately stores the values in each bin as a perfect spatial hash table. To answer a query, we first determine whether or not a record satisfies the query conditions based on the bin boundaries. For the bins with records that can not be resolved, we examine the spatial hash tables. The procedures for examining the bin numbers and the spatial hash tables offer the maximum possible level of concurrency; all records are able to be evaluated by our procedure independently in parallel. Additionally, our Bin-Hash procedures access much smaller amounts of data than similar parallel methods, such as the projection index. This smaller data footprint is critical for certain parallel processors, like GPUs, where memory resources are limited. To demonstrate the effectiveness of Bin-Hash, we implement it on a GPU using the data-parallel programming language CUDA. The concurrency offered by the Bin-Hash index allows us to fully utilize the GPU’s massive parallelism in our work; over 12,000 records can be simultaneously evaluated at any one time. We show that our new query processing method is an order of magnitude faster than current state-of-the-art CPU-based indexing technologies. Additionally, we compare our performance to existing GPU-based projection index strategies. I.

Abstract not found

Abstract. The multi-core trend in CPUs and general purpose graphics processing units (GPUs) offers new opportunities for the database community. The increase of cores at exponential rates is likely to affect virtually every server and client in the coming decade, and presents database management systems with a huge, compelling disruption that will radically change how processing is done. This paper presents a new parallel indexing data structure for answering queries that takes full advantage of the increasing thread-level parallelism emerging in multi-core architectures. In our approach, our Data Parallel Bin-based Index Strategy (DP-BIS) first bins the base data, and then partitions and stores the values in each bin as a separate, bin-based data cluster. In answering a query, the procedures for examining the bin numbers and the bin-based data clusters offer the maximum possible level of concurrency; each record is evaluated by a single thread and all threads are processed simultaneously in parallel. We implement and demonstrate the effectiveness of DP-BIS on two multicore architectures: a multi-core CPU and a GPU. The concurrency afforded by

Typical algorithms for decentralized data distribution work best in a system that is fully built before it first used; adding or removing components results in either extensive reorganization of data or load imbalance in the system. We have developed a decentralized algorithm, RUSH (Replication Under Scalable Hashing), that maps replicated objects to a scalable collection of storage servers or disks. RUSH distributes objects to servers evenly, redistributing as few objects as possible when new servers are added or existing servers are removed to preserve this balanced distribution. It guarantees that replicas of a particular object are not placed on the same server, and allows servers to have different “weights,” distributing more objects to servers with higher weights. The algorithm is very fast, and scales with the number of server groups added to the system. Because there is no central directory, clients can compute data locations in parallel, allowing thousands of clients to access objects on thousands of servers simultaneously.