Engineering topics

Katz, Randy H.

Publications and source records attributed to Katz, Randy H..

14 records

NASA NTRS1993

Interfacing a high performance disk array file server to a Gigabit LAN

Our previous prototype, RAID-1, identified several bottlenecks in typical file server architectures. The most important bottleneck was the lack of a high-bandwidth path between disk, memory, and the network. Workstation servers, such as the Sun-4/280, have very slow access to peripherals on busses far from the CPU. For the RAID-2 system, we addressed this problem by designing a crossbar interconnect, Xbus board, that provides a 40MB/s path between disk, memory, and the network interfaces. However, this interconnect does not provide the system CPU with low latency access to control the various interfaces. To provide a high data rate to clients on the network, we were forced to carefully and efficiently design the network software. A block diagram of the system hardware architecture is given. In the following subsections, we describe pieces of the RAID-2 file server hardware that had a significant impact on the design of the network interface.

Seshan, Srinivasan↗

NASA NTRS1993

RAMA: A file system for massively parallel computers

This paper describes a file system design for massively parallel computers which makes very efficient use of a few disks per processor. This overcomes the traditional I/O bottleneck of massively parallel machines by storing the data on disks within the high-speed interconnection network. In addition, the file system, called RAMA, requires little inter-node synchronization, removing another common bottleneck in parallel processor file systems. Support for a large tertiary storage system can easily be integrated in lo the file system; in fact, RAMA runs most efficiently when tertiary storage is used.

Miller, Ethan L.↗

NASA NTRS1993

The performance of disk arrays in shared-memory database machines

In this paper, we examine how disk arrays and shared memory multiprocessors lead to an effective method for constructing database machines for general-purpose complex query processing. We show that disk arrays can lead to cost-effective storage systems if they are configured from suitably small formfactor disk drives. We introduce the storage system metric data temperature as a way to evaluate how well a disk configuration can sustain its workload, and we show that disk arrays can sustain the same data temperature as a more expensive mirrored-disk configuration. We use the metric to evaluate the performance of disk arrays in XPRS, an operational shared-memory multiprocessor database system being developed at the University of California, Berkeley.

Katz, Randy H.↗

NASA NTRS1993

Diskless supercomputers: Scalable, reliable I/O for the Tera-Op technology base

Computing is seeing an unprecedented improvement in performance; over the last five years there has been an order-of-magnitude improvement in the speeds of workstation CPU's. At least another order of magnitude seems likely in the next five years, to machines with 500 MIPS or more. The goal of the ARPA Teraop program is to realize even larger, more powerful machines, executing as many as a trillion operations per second. Unfortunately, we have seen no comparable breakthroughs in I/O performance; the speeds of I/O devices and the hardware and software architectures for managing them have not changed substantially in many years. We have completed a program of research to demonstrate hardware and software I/O architectures capable of supporting the kinds of internetworked 'visualization' workstations and supercomputers that will appear in the mid 1990s. The project had three overall goals: high performance, high reliability, and scalable, multipurpose system.

Katz, Randy H.↗

NASA NTRS1992

An analysis of file migration in a UNIX supercomputing environment

The super computer center at the National Center for Atmospheric Research (NCAR) migrates large numbers of files to and from its mass storage system (MSS) because there is insufficient space to store them on the Cray supercomputer's local disks. This paper presents an analysis of file migration data collected over two years. The analysis shows that requests to the MSS are periodic, with one day and one week periods. Read requests to the MSS account for the majority of the periodicity; as write requests are relatively constant over the course of a week. Additionally, reads show a far greater fluctuation than writes over a day and week since reads are driven by human users while writes are machine-driven.

Miller, Ethan L.↗

NASA NTRS1992

High-performance network and channel based storage

In the traditional mainframe-centered view of a computer system, storage devices are coupled to the system through complex hardware subsystems called I/O channels. With the dramatic shift toward workstation-based computing, and its associated client/server model of computation, storage facilities are now found attached to file servers and distributed throughout the network. In this paper, we discuss the underlying technology trends that are leading to high-performance network-based storage, namely advances in networks, storage devices, and I/O controller and server architectures. We review several commercial systems and research prototypes that are leading to a new approach to high-performance computing based on network-attached storage.

Katz, Randy H.↗

NASA NTRS1992

Striped tape arrays

A growing number of applications require high capacity, high throughput tertiary storage systems. How data striping ideas apply to arrays of magnetic tape drives is investigated. Data striping increases throughput and reduces response time for large accesses to a storage system. Striped magnetic tape systems are particularly appealing because many inexpensive magnetic tape drives have low bandwidth; striping may offer dramatic performance improvements for these systems. There are several important issues in designing striped tape systems: the choice of tape drives and robots, whether to stripe within or between robots, and the choice of the best scheme for distributing data on cartridges. One of the most troublesome problems in striped tape arrays is the synchronization of transfers across tape drives. Another issue is how improved devices will affect the desirability of striping in the future. The results of simulations comparing the performance of striped tape systems to non-striped systems are presented.

Drapeau, Ann L.↗

NASA NTRS1991

An analytic performance model of disk arrays and its application

As disk arrays become widely used, tools for understanding and analyzing their performance become increasingly important. In particular, performance models can be invaluable in both configuring and designing disk arrays. Accurate analytic performance models are desirable over other types of models because they can be quickly evaluated, are applicable under a wide range of system and workload parameters, and can be manipulated by a range of mathematical techniques. Unfortunately, analytical performance models of disk arrays are difficult to formulate due to the presence of queuing and fork-join synchronization; a disk array request is broken up into independent disk requests which must all complete to satisfy the original request. We develop, validate, and apply an analytic performance model for disk arrays. We derive simple equations for approximating their utilization, response time, and throughput. We then validate the analytic model via simulation and investigate the accuracy of each approximation used in deriving the analytical model. Finally, we apply the analytical model to derive an equation for the optimal unit of data striping in disk arrays.

Lee, Edward K.↗

NASA NTRS1991

Analyzing the I/O behavior of supercomputer applications

Supercomputer applications of I/Os are classified here into three categories - required, checkpoint, and data staging - and it is shown how memory size and CPU speed are likely to affect each category. A data analysis shows that data staging I/O dominates when it is present. If an application requires data staging I/O, its read/write ratio is greater than one; otherwise, its read/write ratio is less than one. This information can be used to anticipate a program's I/O demands and thus reduce the load on the I/O system.

Miller, Ethan L.↗

NASA NTRS1991

High performance network and channel-based storage

In the traditional mainframe-centered view of a computer system, storage devices are coupled to the system through complex hardware subsystems called input/output (I/O) channels. With the dramatic shift towards workstation-based computing, and its associated client/server model of computation, storage facilities are now found attached to file servers and distributed throughout the network. We discuss the underlying technology trends that are leading to high performance network-based storage, namely advances in networks, storage devices, and I/O controller and server architectures. We review several commercial systems and research prototypes that are leading to a new approach to high performance computing based on network-attached storage.

Katz, Randy H.↗

NASA NTRS1991

Robo-line storage: Low latency, high capacity storage systems over geographically distributed networks

Rapid advances in high performance computing are making possible more complete and accurate computer-based modeling of complex physical phenomena, such as weather front interactions, dynamics of chemical reactions, numerical aerodynamic analysis of airframes, and ocean-land-atmosphere interactions. Many of these 'grand challenge' applications are as demanding of the underlying storage system, in terms of their capacity and bandwidth requirements, as they are on the computational power of the processor. A global view of the Earth's ocean chlorophyll and land vegetation requires over 2 terabytes of raw satellite image data. In this paper, we describe our planned research program in high capacity, high bandwidth storage systems. The project has four overall goals. First, we will examine new methods for high capacity storage systems, made possible by low cost, small form factor magnetic and optical tape systems. Second, access to the storage system will be low latency and high bandwidth. To achieve this, we must interleave data transfer at all levels of the storage system, including devices, controllers, servers, and communications links. Latency will be reduced by extensive caching throughout the storage hierarchy. Third, we will provide effective management of a storage hierarchy, extending the techniques already developed for the Log Structured File System. Finally, we will construct a protototype high capacity file server, suitable for use on the National Research and Education Network (NREN). Such research must be a Cornerstone of any coherent program in high performance computing and communications.

Katz, Randy H.↗

NASA NTRS1991

Performance of a disk array prototype

A prototype disk array's component-performance limits are examined using SCSI bus traces, Sprite operating system traces, and user programs. The array performs successfully for a workload of small, random I/O operations, achieving 275 I/Os per second on 14 disks before the Sun4/280 host becomes CPU-limited. The prototype is less successful in delivering high throughput for large, sequential operations. Memory system contention on the Sun4/280 host limits throughput to 2.3 MBytes/sec under the Sprite Operating System. Throughput is also limited by the bandwidth supported by the VME backplane, disk controller and disks, and overheads associated with the SCSI protocol. We conclude that merely using a powerful host CPU and many disks will not provide the full bandwidth possible from disk arrays. Host memory bandwidth and throughput of disk controllers are equally important. In addition, operating systems should avoid unnecessary copy and cache flush operations that can saturate the host memory system.

Chervenak, Ann L.↗

NASA NTRS1991

Performance consequences of parity placement in disk arrays

The performance of a variety of parity placement schemes are defined and investigated to demonstrate that, at relatively large request sizes of hundreds of kilobytes, the choice of parity placement significantly affects performance (20 to 30 percent for the disk array configurations that are common today). It is shown that the left-symmetric, extended-left-symmetric and flat-left-symmetric are the best RAID level 5 parity placements. The placement with the highest read performance, flat-left-symmetric, has the lowest write performance, while the placement with the lowest read performance, left-symmetric, has the highest write performance. Suggestions for optimizing parity placements are included.

Lee, Edward K.↗

NASA NTRS1988

A case for Redundant Arrays of Inexpensive Disks (RAID)

Increasing performance of CPUs and memories will be squandered if not matched by a similar performance increase in I/O. While the capacity of Single Large Expensive Disks (SLED) has grown rapidly, the performance improvement of SLED has been modest. Redundant Arrays of Inexpensive Disks (RAID), based on the magnetic disk technology developed for personal computers, offers an attractive alternative to SLED, promising improvements of an order of magnitude in performance, reliability, power consumption, and scalability. This paper introduces five levels of RAIDs, giving their relative cost/performance, and compares RAID to an IBM 3380 and a Fujitsu Super Eagle.

Patterson, David A.↗