We present a new technique for simulating fair coin flips using a biased, stationary source of randomness. Sequences of random numbers are of pervasive importance in cryptography and vital to many other computing applications. Many sources of randomness, such as radioactive or quantum-mechanical sources, possess the property of stationarity. In other words, they produce independent outputs over fixed probability distributions. The output of such sources may be viewed as the result of rolling a biased or loaded die. While a biased die may be a good source of entropy, many applications require input in the form of unbiased bits, rather than biased ones. For this reason, von Neumann (1951) presented a now well-known and extensively investigated technique for using a biased coin to simulate a fair coin. We describe a new generalization of von Neumann's algorithm distinguished by its high level of practicality and amenability to analysis. In contrast to previous efforts, we are able to prove our algorithm optimally efficient, in the sense that it simulates the maximum possible number of fair coin flips for a given number of die rolls. In fact, we are able to prove that in an asymptotic sense our algorithm extracts the full entropy of its input. Moreover, we demonstrate experimentally that our algorithm achieves a high level of computational and output efficiency in a practical setting.

An increasing number of database applications require online access to massive amounts of data. Since large scale storage systems implemented entirely on magnetic disk can be impractical or too costly for many applications, tape jukeboxes can provide an attractive solution. The paper shows how the performance of tape jukeboxes can be improved across a broad parameter space via a new scheduling algorithm and schemes for the placement and replication of hot data. We substantiate our claim by an extensive simulation study that quantifies the improvements obtained over a wide variety of workload characteristics. Our experiments suggest that system throughput increases when replicas of hot data are placed at the tape ends (not in the middle or at the beginning). As a result, the proposed replication techniques can be used to fill existing spare capacity in a tape jukebox, thus improving the performance of the jukebox "for free".

In modern I/O architectures, multiple disk drives are attached to each I/O controller.A study of the performance of such architectures under I/O-intensive workloads has revealed a performance impairment that results from a previously unknown form of convoy behavior in disk I/O.In this paper, we describe measurements of the read performance of multiple disks that share a SCSI bus under a heavy workload, and develop and validate formulas that accurately characterize the observed performance (to within 12% on several platforms for I/O sizes in the range 16-128 KB).Two terms in the formula clearly characterize the lost performance seen in our experiments.We describe techniques to deal with the performance impairment, via user-level workarounds that achieve greater overlap of bus transfers with disk seeks, and that increase the percentage of transfers that occur at the full bus bandwidth rather than at the lower bandwidth of a disk head.Experiments show bandwidth improvements of lo-20% when using these user-level techniques, but only in the case of large I/OS.

We sugg=t a practical and economical way to generate random bits using a computer disk drive * a source of randomn-.It requirw no additiond hardware (given a system with a disk), and no user involvement.As a concrete example of performance, on a Sun Wtra-1 with a Seagate Cheetah disk, it generatw bits at a rate of either 5 bits per minute or 577 bits per minute depending on the physical phenomena that we use = a source of randomness.The generated bits are random by a theoretical argument, and *O pass a severe battery of statiaticrd twts. 'InformationSciences Research Center, Bell Laboratories.{m=hsj ,sbriver,bruce} arese=.h.bell-labs .com

For a wide variety of computational tasks, disk I/O continues to be a serious obstacle to high performance. The focus of the present paper is on systems that use multiple disks per SCSI bus. We measured the performance of concurrent random I/Os, and observed bus-related phenomena that impair performance. We describe these phenomena, and present a new I/O performance model that accurately predicts the average bandwidth achieved by a heavy workload of random reads from disks on a SCSI bus. This model, although relatively simple, predicts performance on several platforms to within 12% for I/O sizes in the range 16-128 KB. We describe a technique to improve the I/O bandwidth by 10-20% for random-access workloads that have large I/Os and high concurrency. This technique increases the percentage of disk head positioning time that is overlapped with data transfers, and increases the percentage of transfers that occur at bus bandwidth, rather than at disk-head bandwidth.

New applications require online access to many terabytes of data, but a magnetic disk storage system this large requires thousands of drives. Magnetic tape is be a good alternative, except that the application demand for transparent data retrieval is not met by current tape systems because of their high access latency. This latency can be significantly improved by good retrieval scheduling. A fundamental prerequisite to efficient scheduling is the ability to estimate the amount of time required for tape positioning operations (the locate time). For serpentine tape, which is the most common mass storage tape technology, this estimation is subtle and complex. The main contribution of this paper is a locate-time model for a DLT4000 tape drive. The accuracy of the model is evaluated by measurements, and the utility of the model is demonstrated through a model-driven simulation of retrieval scheduling, validated by measurements and sensitivity testing. In brief, the locate-time model is accurate to within a few percent, which enables the production of efficient schedules.

New database applications that require the storage and retrieval of many terabytes of data are reaching the limits for disk-based storage systems, in terms of both cost and scalability. These limits provide a strong incentive for the development of databases that augment disk storage with technologies better suited to large volumes of data. In particular, the seamless incorporation of tape storage into database systems would be of great value. Tape storage is two orders of magnitude more efficient than disk in terms of cost per terabyte and physical volume per terabyte; however, a key problem is that the random access latency of tape is three to four orders of magnitude slower than disk. Thus, to incorporate a tape bulk store in an online storage system, the problem of tape access latency must be solved. One approach to reducing the latency is careful I/O scheduling. The focus of this paper is on efficient random I/O scheduling for tape drives that use a serpentine track layout, such as the Quantum DLT and the IBM 3480 and 3590. For serpentine tape, I/O scheduling is problematic because of the complex relationships between logical block numbers, their physical positions on tape, and the time required for tape positioning between these physical positions. The results in this paper show that our scheduling schemes provide a significant improvement in the latency of random access to serpentine tape.

In recent years, the development of expert systems implemented by rule-based production systems has emerged as one of the dominant paradigms in the field of artificial intelligence. While production systems offer important advantages in large-scale AI applications, their use in such applications is typically very costly in execution time. In this paper, we describe an algorithm for executing production systems expressed in the OPS5 language on a massively parallel multiple-SIMD machine called NON-VON, portions of which are currently under construction at Columbia University. The algorithm, a parallel adaptation of Forgy's Rete Match, has been implemented and tested on an instruction-level simulator. We present a detailed performance analysis, based on the implemented code, for the averaged characteristics of six production systems having an average of 910 inference rules each. The analysis predicts an execution rate of more than 850 production firings per second using hardware comparable in cost to a VAX 11/780. By way of comparison, a LISP-based OPS5 interpreter running on a VAX 11/780 typically fires 1 to 5 rules per second, while a Bliss-based interpreter executes 5 to 12 rules per second.

In a paper by Hawthorn and DeWitt, the projected performance of several proposed database machines was examined for three relational database queries. The present paper investigates the performance of a massively parallel machine called NON-VON for the same queries under comparable assumptions. In the case of simple queries, a NON-VON machine of comparable size to those considered by Hawthorn and DeWitt is found to be somewhat faster than the fastest machines examined in their study; for a more complex database operation, NON-VON is shown to be five to ten times faster than the fastest of these machines.