Browsing by Author "Selvitopi, R. O."

Now showing 1 - 3 of 3

Open Access
A novel method for scaling iterative solvers: avoiding latency overhead of parallel sparse-matrix vector multiplies
(Institute of Electrical and Electronics Engineers, 2015) Selvitopi, R. O.; Ozdal, M. M.; Aykanat, Cevdet
In parallel linear iterative solvers, sparse matrix vector multiplication (SpMxV) incurs irregular point-to-point (P2P) communications, whereas inner product computations incur regular collective communications. These P2P communications cause an additional synchronization point with relatively high message latency costs due to small message sizes. In these solvers, each SpMxV is usually followed by an inner product computation that involves the output vector of SpMxV. Here, we exploit this property to propose a novel parallelization method that avoids the latency costs and synchronization overhead of P2P communications. Our method involves a computational and a communication rearrangement scheme. The computational rearrangement provides an alternative method for forming input vector of SpMxV and allows P2P and collective communications to be performed in a single phase. The communication rearrangement realizes this opportunity by embedding P2P communications into global collective communication operations. The proposed method grants a certain value on the maximum number of messages communicated regardless of the sparsity pattern of the matrix. The downside, however, is the increased message volume and the negligible redundant computation. We favor reducing the message latency costs at the expense of increasing message volume. Yet, we propose two iterative-improvement-based heuristics to alleviate the increase in the volume through one-to-one task-to-processor mapping. Our experiments on two supercomputers, Cray XE6 and IBM BlueGene/Q, up to 2,048 processors show that the proposed parallelization method exhibits superior scalable performance compared to the conventional parallelization method.
Open Access
Replicated partitioning for undirected hypergraphs
(Academic Press, 2012) Selvitopi, R. O.; Turk, A.; Aykanat, Cevdet
Hypergraph partitioning (HP) and replication are diverse but powerful tools that are traditionally applied separately to minimize the costs of parallel and sequential systems that access related data or process related tasks. When combined together, these two techniques have the potential of achieving significant improvements in performance of many applications. In this study, we provide an approach involving a tool that simultaneously performs replication and partitioning of the vertices of an undirected hypergraph whose vertices represent data and nets represent task dependencies among these data. In this approach, we propose an iterative-improvement-based replicated bipartitioning heuristic, which is capable of move, replication, and unreplication of vertices. In order to utilize our replicated bipartitioning heuristic in a recursive bipartitioning framework, we also propose appropriate cut-net removal, cut-net splitting, and pin selection algorithms to correctly encapsulate the two most commonly used cutsize metrics. We embed our replicated bipartitioning scheme into the state-of-the-art multilevel HP tool PaToH to provide an effective and efficient replicated HP tool, rpPaToH. The performance of the techniques proposed and the tools developed is tested over the undirected hypergraphs that model the communication costs of parallel query processing in information retrieval systems. Our experimental analysis indicates that the proposed technique provides significant improvements in the quality of the partitions, especially under low replication ratios. © 2012 Elsevier Inc. All rights reserved.
Open Access
Temporal workload-aware replicated partitioning for social networks
(Institute of Electrical and Electronics Engineers, 2014-11) Turk, A.; Selvitopi, R. O.; Ferhatosmanoglu, H.; Aykanat, Cevdet
Most frequent and expensive queries in social networks involve multi-user operations such as requesting the latest tweets or news-feeds of friends. The performance of such queries are heavily dependent on the data partitioning and replication methodologies adopted by the underlying systems. Existing solutions for data distribution in these systems involve hash- or graph-based approaches that ignore the multi-way relations among data. In this work, we propose a novel data partitioning and selective replication method that utilizes the temporal information in prior workloads to predict future query patterns. Our method utilizes the social network structure and the temporality of the interactions among its users to construct a hypergraph that correctly models multi-user operations. It then performs simultaneous partitioning and replication of this hypergraph to reduce the query span while respecting load balance and I/O load constraints under replication. To test our model, we enhance the Cassandra NoSQL system to support selective replication and we implement a social network application (a Twitter clone) utilizing our enhanced Cassandra. We conduct experiments on a cloud computing environment (Amazon EC2) to test the developed systems. Comparison of the proposed method with hash- and enhanced graph-based schemes indicate that it significantly improves latency and throughput.