In a path-breaking paper last year Pat and Betty O'Neil and Gerhard Weikum proposed a self-tuning improvement to the Least Recently Used (LRU) buffer management algorithm[15]. Their improvement is called LRU/k and advocates giving priority to buffer pages based on the kth most recent access. (The standard LRU algorithm is denoted LRU/1 according to this terminology.) If P1's kth most recent access is more more recent than P2's, then P1 will be replaced after P2. Intuitively, LRU/k for k ? 1 is a good strategy, because it gives low priority to pages that have been scanned or to pages that belong to a big randomly accessed file (e.g., the account file in TPC/A). They found that LRU/2 achieves most of the advantage of their method. The one problem of LRU/2 is the processor Supported by U.S. Office of Naval Research #N00014-91-J1472 and #N00014-92-J-1719, U.S. National Science Foundation grants #CCR-9103953 and IRI-9224601, and USRA #5555-19. Part of this work was performed while Theodo...

This paper reports our research results that improve second level buer cache performance. Several previous studies have shown that a good single level cache replacement algorithm such as LRU does not work well with second level buer caches. Second level buer caches have dierent access pattern from rst level buer caches because Accesses to second level buer caches are actually misses from rst level buer caches.

Improvements in the processing speed of multiprocessors are outpacing improvements in the speed of disk hardware. Parallel disk I/O subsystems have been proposed as one way to close the gap between processor and disk speeds. In a previous paper we showed that prefetching and caching have the potential to deliver the performance bene ts of parallel le systems to parallel applications. In this paper we describe experiments with practical prefetching policies, and show that prefetching can be implemented e ciently even for the more complex parallel le access patterns. We also test the ability of these policies across a range of architectural parameters. 1

In an on-demand video server environment, multimedia objects (e.g. movies) are very large and are read sequentially. Hence it is not economical to cache the entire object. However, caching random fractions of a multimedia object is not beneficial. This is due to the stringent response time requirements where continuous availability of a stream has to be guaranteed; whereas caching random fractions will result in unpredictable load on the disks. Therefore, traditional buffer management policies such as LRU are not effective. In addition, the sequential access implies pages brought in by a stream can be reused by a closely following stream and subsequently discarded, thus buffering only a fraction of the entire object. In this paper, we propose a buffer management policy called the interval caching policy based on the above idea that identifies certain streams and temporarily buffers the pages brought in by those streams. We study the efficacy of this technique for reducing disk overload...

The timeliness and synchronization requirement of multimedia data demands efficient buffer management and disk access schemes for multimedia database systems (MMDBS). The data rates involved are also very high and despite the development of efficient storage and retrieval strategies, disk I/O is likely to be a bottleneck, thereby limiting the number of concurrent sessions supported by a system. This calls for better use of data that has already been brought into the buffer by exploiting sharing whenever possible using advance knowledge of the multimedia stream to be accessed. This paper introduces the notion of continuous media caching which is a simple and novel technique where buffers that have been played back by a user are preserved in a controlled fashion for use by subsequent users requesting the same data. This is shown to have considerable impact on the performance of buffer management schemes. When continuous media sharing is used in conjunction with batching of user requests...

CLOCK is a classical cache replacement policy dating back to 1968 that was proposed as a low-complexity approximation to LRU. On every cache hit, the policy LRU needs to move the accessed item to the most recently used position, at which point, to ensure consistency and correctness, it serializes cache hits behind a single global lock. CLOCK eliminates this lock contention, and, hence, can support high concurrency and high throughput environments such as virtual memory (for example, Multics, UNIX, BSD, AIX) and databases (for example, DB2). Unfortunately, CLOCK is still plagued by disadvantages of LRU such as disregard for "frequency", susceptibility to scans, and low performance. As our main contribution, we propose a simple and elegant new algorithm, namely, CLOCK with Adaptive Replacement (CAR), that has several advantages over CLOCK: (i) it is scan-resistant; (ii) it is self-tuning and it adaptively and dynamically captures the "recency" and "frequency" features of a workload; (iii) it uses essentially the same primitives as CLOCK, and, hence, is low-complexity and amenable to a high-concurrency implementation; and (iv) it outperforms CLOCK across a wide-range of cache sizes and workloads. The algorithm CAR is inspired by the Adaptive Replacement Cache (ARC) algorithm, and inherits virtually all advantages of ARC including its high performance, but does not serialize cache hits behind a single global lock. As our second contribution, we introduce another novel algorithm, namely, CAR with Temporal filtering (CART), that has all the advantages of CAR, but, in addition, uses a certain temporal filter to distill pages with long-term utility from those with only short-term utility.

Learning from experience to predict sequences of discrete symbols is a fundamental problem in machine learning with many applications. We present a simple and practi-ca] algorithm (TDAG) for discrete sequence prediction. Based on a text-compression method, the TDAG algorithm limits the growth of storage by retaining the most likely prediction contexts and discarding (forgetting) less likely ones. The storage/speed tradeoffs are parameterized so that the algorithm can be used in a variety of applications. Our experiments verify its performance on data compression tasks and show how it applies to two problems: dynamica]ly optimizing Prolog programs for good average-case behavior and maintaining a cache for a database on mass storage.

systems. These correlations can be exploited for improving the effectiveness of storage caching, prefetching, data layout and disk scheduling. Unfortunately, information about block correlations is not available at the storage system level. Previous approaches for discovering file correlations in file systems do not scale well enough to be used for discovering block correlations in storage systems.

Abstract. The analytic prediction of buffer hit probability, based on the charac-terization of database accesses from real reference traces, is extremely useful for workload management and system capacity planning. The knowledge can be help-ful for proper allocation of buffer space to various database relations, as well as for the management of buffer space for a mixed transaction and query environment. Access characterization can also be used to predict the buffer invalidation effect in a multi-node environment which, in turn, can influence transaction routing strate-gies. However, it is a challenge to characterize the database access pattern of a real workload reference trace in a simple manner that can easily be used to compute buffer hit probability. In this article, we use a characterization method that distin-guishes three types of access patterns from a trace: (1) locality within a transaction, (2) random accesses by transactions, and (3) sequential accesses by long queries. We then propose a concise way to characterize the access skew across randomly accessed pages by logically grouping the large number of data pages into a small number of partitions such that the frequency of accessing each page within a par-tition can be treated as equal. Based on this approach, we present a recursive binary partitioning algorithm that can infer the access skew characterization from the buffer hit probabilities for a subset of the buffer sizes. We validate the buffer hit predictions for single and multiple node systems using production database traces. We further show that the proposed approach can predict the buffer hit probability of a composite workload from those of its component files. Key Words. Database access characterization, access skew, sequential access, ref-erence trace, workload management, analytic prediction.

As improvements in processor performance continue to far outpace improvements in storage performance, I /O is increasingly the bottleneck in computer systems, especially in large database systems that manage huge amounts of data. The key to achieving good I /O performance is to thoroughly understand its characteristics. In this article we present a comprehensive analysis of the logical I/O reference behavior of the peak production database workloads from ten of the world’s largest corporations. In particular, we focus on how these workloads respond to different techniques for caching, prefetching, and write buffering. Our findings include several broadly applicable rules of thumb that describe how effective the various I /O optimization techniques are for the production workloads. For instance, our results indicate that the buffer pool miss ratio tends to be related to the ratio of buffer pool size to data size by an inverse square root rule. A similar fourth root rule relates the write miss ratio and the ratio of buffer pool size to data size. In addition, we characterize the reference characteristics of workloads similar to the Transaction Processing Performance Council (TPC) benchmarks C (TPC-C) and D (TPC-D), which are de facto standard performance measures for online transaction processing (OLTP) systems and decision support systems (DSS), respectively. Since benchmarks such as TPC-C and TPC-D can only be