Browsing by Author "Kallahalla, Mahesh"

Now showing 1 - 7 of 7
  • Thumbnail Image
    Item
    Analysis of simple randomized buffer management for parallel I/O
    (Elsevier, 2004-04) Kallahalla, Mahesh; Varman, Peter J.; Electrical and Computer Engineering; Varman Laboratory
    Buffer management for a D-disk parallel I/O system is considered in the context of randomized placement of data on the disks. A simple prefetching and caching algorithm PHASE-LRU using bounded lookahead is described and analyzed. It is shown that PHASE-LRU performs an expected number of I/Os that is within a factor (logD/log logD) of the number performed by an optimal off-line algorithm. In contrast, any deterministic buffer
  • Thumbnail Image
    Item
    ASP: Adaptive Online Parallel Disk Scheduling
    (1999) Kallahalla, Mahesh; Varman, Peter J.
    In this work we address the problems of prefetching and I/O scheduling for read-once reference strings in a parallel I/O system. We use the standard parallel disk model with D disks a shared I/O bu er of sizeM. We design an on-line algorithm ASP (Adaptive Segmented Prefetching) with ML-block lookahead, L 1, and compare its performance to the best on-line algorithm with the same lookahead. We show that for any reference string the number of I/Os done by ASP is with a factor (C), C = minfpL;D1=3g, of the number of I/Os done by the optimal algorithm with the same amount of lookahead.
  • Thumbnail Image
    Item
    Competitive Parallel Disk Prefetching and Buffer Management1
    (Rice University, 2000) Barve, Rakesh; Kallahalla, Mahesh; Varman, Peter J.
    We provide a competitive analysis framework for online prefetching and buffer management algorithms in parallel IrO systems, using a read-once model of block references. This has widespread applicability to key IrO-bound applications such as external merging and concurrent playback of multiple video streams. Two realistic lookahead models, global lookahead and local lookahead, are defined. Algorithms NOM and GREED, based on these two forms of lookahead are analyzed for shared buffer and distributed buffer configurations, both of which occur frequently in existing systems. An important aspect of our work is that we show how to implement both of the models of lookahead in practice using the simple techniques of forecasting and flushing. Given a D-disk parallel IrO system and a globally shared IrO buffer that can hold up to M disk blocks, we derive a lower bound of V 'D . on the competitive ratio of any deterministic online prefetching algorithm with O M. lookahead. NOM is shown to match the lower bound using global M-block lookahead. In contrast, using only local lookahead results in an V D. competitive ratio. When the buffer is distributed into D portions of MrD blocks each, the algorithm GREED based on local lookahead is shown to be optimal, and NOM is within a constant factor of optimal. Thus we provide a theoretical basis for the intuition that global lookahead is more valuable for prefetching in the case of a shared buffer configuration, whereas it is enough to provide local lookahead in the case of a distributed configuration. Finally, we analyze the performance of these algorithms for reference strings generated by a uniformly-random stochastic process and we show that they achieve the minimal expected number of IrOs. These results also give bounds on the worst-case expected performance of algorithms which employ randomization in the data layout. 1 A preliminary version of this paper has appeared in the Proceedings of the ACM Fifth Annual Workshop on IrO in Parallel and Distributed Systems. 2 Supported in part by an IBM graduate fellowship. Part of this work was done while the author was visiting Lucent Technologies, Bell Laboratories, Murray Hill, NJ. 3 Supported in part by a grant from the Schlumberger Foundation and by the National Science Foundation under Grant CCR-9704562. 4 Supported in part by the National Science Foundation under Grant CCR-9522047 and by Army Research Office MURI Grant DAAH04-96-1-0013. Part of this work was done while the author was visiting Lucent Technologies, Bell Laboratories, Murray Hill, NJ.
  • Thumbnail Image
    Item
    Competitive prefetching and buffer management for parallel I/O systems
    (1997) Kallahalla, Mahesh; Varman, Peter J.
    In this thesis we study prefetching and buffer management algorithms for parallel I/O systems. Two models of lookahead, global and local, which give limited information regarding future accesses are introduced. Two configurations of the I/O buffer, shared and distributed, are considered, based upon the accessibility of the I/O buffer. The performance of prefetching algorithms using the two forms of lookahead is analyzed in the framework of competitive analysis, for read-once access patterns. Two algorithms, PHASE and GREED, which match the lower bounds are presented. A randomized version of GREED that performs the minimal expected number of I/Os is designed and applied to the problems of external sorting and video retrieval. Finally the problem of designing prefetching and buffer management algorithms for read-many reference strings is examined. An algorithm which uses randomized write-back to attain good expected I/O performance is presented.
  • Thumbnail Image
    Item
    PC-OPT: Optimal Offline Prefetching and Caching for Parallel I/O Systems
    (IEEE Xplore, 2002-11) Kallahalla, Mahesh; Varman, Peter J.; Electrical and Computer Engineering; Varman Laboratory
    Abstract—We address the problem of prefetching and caching in a parallel I/O system and present a new algorithm for parallel disk scheduling. Traditional buffer management algorithms that minimize the number of block misses are substantially suboptimal in a parallel I/O system where multiple I/Os can proceed simultaneously. We show that in the offline case, where a priori knowledge of all the requests is available, PC-OPT performs the minimum number of I/Os to service the given I/O requests. This is the first parallel I/O scheduling algorithm that is provably offline optimal in the parallel disk model. In the online case, we study the context of global L-block lookahead, which gives the buffer management algorithm a lookahead consisting of L distinct requests. We show that the competitive ratio of PC-OPT, with global L-block lookahead, is M ÿ L D , when L M, and MD=L , when L > M, where the number of disks is D and buffer size is M.
  • Thumbnail Image
    Item
    Prefetching and buffer management for parallel I/O systems
    (2000) Kallahalla, Mahesh; Varman, Peter J.
    In parallel I/O systems the I/O buffer can be used to improve I/O parallelism by improving I/O latency by caching blocks to avoid repeated disk accesses for the same block, and also by buffering prefetched blocks and making the load on disks more uniform. To make best use of available parallelism and locality in I/O accesses, it is necessary to design prefetching and caching algorithms that schedule reads intelligently so that the most useful blocks are prefetched into the buffer and the most valuable blocks are retained in the buffer when the need for evictions arises. This dissertation focuses on algorithms for buffer management in parallel I/O systems. Our aim is to exploit the high parallelism provided by multiple disks to reduce the average read latency seen by an application. The thesis is that traditional greedy strategies fail to exploit I/O parallelism thereby necessitating new algorithms to make use of the available I/O resources. We show that buffer management in parallel I/O systems is fundamentally different from that in systems with a single disk, and develop new algorithms that carefully decide which blocks to prefetch and when, together with which blocks to retain in the buffer. Our emphasis is on designing computationally simple algorithms for optimizing the number of I/Os performed. We consider two classes of I/O access patterns, read-once and read-often, based on the frequency of accesses to the same data. With respect to buffer management for both classes of accesses, we identify fundamental bounds on performance of online algorithms, study the performance of intuitive strategies, and present randomized and deterministic algorithms that guarantee higher performance.
  • Thumbnail Image
    Item
    Randomized Parallel Prefetching and Buffer Management
    (Kluwer Academic Publishers, 1999) Kallahalla, Mahesh; Varman, Peter J.
    We show that deterministic algorithms using bounded lookahead cannot fully exploit the potential of a parallel I/O system. Randomization can be used to significantly improve the performance of parallel prefetching and buffer management algorithms. Using randomization in the data layout and a simple prefetching scheme, we show that a readonce reference string of length N can be serviced in θ(N/D) parallel I/Os in a D-disk system. For the case of read-many reference strings we introduce a novel algorithm using randomized write-back with a competitive ratio of θ(D). In contrast, we show that deterministic write-back results in a competitive ratio of at least (D). Supported in part by NSF Grant CCR-9704562 and a grant from the Schlumberger Foundation.