Results 1 
2 of
2
HyCache+: Towards Scalable HighPerformance Caching Middleware for Parallel File Systems
"... Abstract—The evergrowing gap between the computation and I/O is one of the fundamental challenges for future computing systems. This computationI/O gap is even larger for modern large scale highperformance systems due to their stateoftheart yet decades long architecture: the compute and storag ..."
Abstract

Cited by 6 (1 self)
 Add to MetaCart
(Show Context)
Abstract—The evergrowing gap between the computation and I/O is one of the fundamental challenges for future computing systems. This computationI/O gap is even larger for modern large scale highperformance systems due to their stateoftheart yet decades long architecture: the compute and storage resources form two cliques that are interconnected with shared networking infrastructure. This paper presents a distributed storage middleware, called HyCache+, right on the compute nodes, which allows I/O to effectively leverage the high bisection bandwidth of the highspeed interconnect of massively parallel highend computing systems. HyCache+ provides the POSIX interface to end users with the memoryclass I/O throughput and latency, and transparently swap the cached data with the existing slowspeed but highcapacity networked attached storage. HyCache+ has the potential to achieve both high performance and lowcost large capacity, the best of both worlds. To further improve the caching performance from the perspective of the global storage system, we propose a 2phase mechanism to cache the hot data for parallel applications, called 2Layer Scheduling (2LS), which minimizes the file size to be transferred between compute nodes and heuristically replaces files in the cache. We deploy HyCache+ on the IBM BlueGene/P supercomputer, and observe two orders of magnitude faster I/O throughput than the default GPFS parallel file system. Furthermore, the proposed heuristic caching approach shows 29X speedup over the traditional LRU algorithm. Index Terms—distributed caching, parallel and distributed file systems, heterogeneous storage I.
Theoretical Work General Approach
"... Many optimization problems, both of practical and theoretical importance, are NPhard. For NPhard optimization problems, there is no hope of designing efficient (i.e., polynomialtime) algorithms giving the exact solution. One way of dealing with NPHard optimization problems is to find polynomial ..."
Abstract
 Add to MetaCart
(Show Context)
Many optimization problems, both of practical and theoretical importance, are NPhard. For NPhard optimization problems, there is no hope of designing efficient (i.e., polynomialtime) algorithms giving the exact solution. One way of dealing with NPHard optimization problems is to find polynomialtime approximation algorithms. These algorithms are guaranteed to be fast and to return a nearoptimal answer. A typical result is a polynomialtime algorithm for the traveling salesman problem and a proof that the output of the algorithm has value at most 50 % more than the optimum. Besides having a worstcase guarantee, approximation algorithms have two other attractive features. First, often, on real world problems, the approximation algorithms perform much better than the worstcase ratio proved. Sometimes, approximation algorithms are the best performers in practice. Second, to prove a better performance guarantee often requires one to obtain a better understanding of the combinatorial structure of the problem. 1 Scheduling Issues We propose investigating the algorithmic aspects of assigning jobs to compute nodes together with the scheduling the transfer of files between the cache of the processing and the network attached storage, with the objectives of minimizing makespan and/or communication costs.