Xref: world comp.org.usenix:4861
Path: world!news.mathworks.com!hookup!swrinde!sdd.hp.com!hplabs!hplextra!news.dtc.hp.com!col.hp.com!cello.hpl.hp.com!golding
From: [email protected] (Richard Golding)
Newsgroups: comp.org.usenix
Subject: References from Winter Conference
Date: 25 Jan 1995 00:50:39 GMT
Organization: Hewlett Packard Labs, Palo Alto, CA, USA
Lines: 837
Message-ID: <[email protected]>
NNTP-Posting-Host: cello.hpl.hp.com
Keywords: bibliographic references
X-Newsreader: TIN [version 1.2 PL2]

I've scanned in the abstracts and information from the proceedings of
the New Orleans conferences, and formatted them for refdbms.  Please
let me know if you find any mistakes in the entries.

Refdbms users: these will be added to the `usenix' database this
evening and should propagate to your site in the next day or so.

-richard

%z InProceedings
%K Stevens95
%A W. Richard Stevens
%y Consultant
%A Jan-Simon Pendry
%y Sequent UK
%T Portals in 4.4BSD
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 1 10
%x Portals were added to 4.4BSD as an experimental feature and are in
%x the publicly available 4.4BSD-Lite distribution. Portals provide
%x access to alternate file types or devices using names in the normal
%x filesystem that a process just opens. For example, an open of
%x /p/tcp/foo.com/smtp returns a TCP socket descriptor to the calling
%x process that is connected to the SMTP server on the specified
%x host. By providing access through the normal filesystem, the
%x calling process need not be aware of the special functions
%x necessary to create a TCP socket and establish a TCP
%x connection. This makes TCP connections, for example, available to
%x programs such as Awk, Tcl, and shell scripts.
%x \par
%x This paper describes the implementation of portals in 4.4BSD as
%x another type of filesystem and provides some examples.


%z InProceedings
%K John95
%A Aju John
%y Digital Equipment Corporation
%T Dynamic vnodes - design and implementation
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 11 23
%x Dynamic vnodes make the UNIX kernel responsive to a varying demand
%x for vnodes, without a need to rebuild the kernel. It also optimizes
%x the usage of memory by deallocating excess vnodes. This paper
%x describes the design and implementation of dynamic vnodes in DEC
%x OSF/l V3.0. The focus is on the vnode deallocation logic in a
%x Symmetric MultiProcessing environment.
%x \par
%x Deallocation of vnodes differs from the familiar concept of
%x dynamically allocated data structures in the following ways: the
%x legacy name-cache design implicitly assumes that vnodes are never
%x deallocated, and the vnode free-list needs to cache unused vnodes
%x effectively.

%z InProceedings
%K Pendry95
%A Jan-Simon Pendry
%y Sequent UK
%A Marshall Kirk McKusick
%y Author and Consultant
%T Union mounts in 4.4BSD-Lite
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 25 33
%x This paper describes the design and rationale behind union mounts,
%x a new filesystem-namespace management tool available in
%x 4.4BSDLite. Unlike a traditional mount that hides the contents of
%x the directory on which it is placed, a union mount presents a view
%x of a merger of the two directories. Although only the filesystem at
%x the top of the union stack can be modified, the union filesystem
%x gives the appearance of allowing anything to be deleted or
%x modified. Files in the lower layer may be deleted with whiteout in
%x the top layer. Files to be modified are automatically copied to the
%x top layer.
%x \par
%x This new functionality makes possible several new applications
%x including the ability to apply patches to a CD-ROM and eliminate
%x symbolic links generated by an automounter. Also possible is the
%x provision of per-user views of the filesystem, allowing private
%x views of a shared work area, or local builds from a centrally
%x shared read-only source tree.

%z InProceedings
%K Devarakonda95
%A Murthy Devarakonda
%A Ajay Mohindra
%A Jill Simoneaux
%A William H. Tetzlaff
%y IBM Thomas J. Watson Research Center, Hawthorne, NY
%T Evaluation of design alternatives for a cluster file system
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 35 46
%x Based on implementation experience and measurements, this paper
%x presents an evaluation of design alternatives to a cinder file
%x system. The file system is targeted for IBM cluster systems,
%x Scalable POWERparallel and AIX HACMP/6000. We considered a shared
%x disk approach where serialized, multiple instances of a
%x single-system file system directly access file data as disk blocks,
%x and a shared file system approach which is the conventional method
%x of distributing file system function between a client and a
%x server. We conclude that the shared disk approach suffers from the
%x difficulties of metadata serialization, poor write-sharing
%x performance, and read throughput.

%z InProceedings
%K Goldick95
%A Jonathan S. Goldick
%A Kathy Benninger
%A Christopher Kirby
%A Christopher Maher
%A Bill Zumach
%y Pittsburgh Supercomputing Center
%T Multi-resident AFS: an adventure in mass storage
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 47 58
%x The Pittsburgh Supercomputing Center has been working to integrate
%x distributed file system technology with hierarchical mass
%x storage. We produced a system utilizing the Andrew File System that
%x can be interfaced to many mass storage systems. We retained the
%x semantics of AFS and compatibility with standard clients and
%x servers. The architecture has a logical separation between the
%x facility that provides the user interface and access semantics and
%x the management of the storage systems that contain user
%x data. Support for file level replication is provided for high
%x availability to data in a fashion that is transparent to
%x users. This system is called Multi-Resident AFS.

%z InProceedings
%K Miller95
%A Ethan L. Miller
%y University of Maryland Baltimore County
%A Randy H. Katz
%y University of California at Berkeley
%T RAMA: easy access to a high-bandwidth massively parallel file system
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 59 70
%x Massively parallel file systems must provide high bandwidth file
%x access to programs running on their machines. Most accomplish this
%x goal by striping files across arrays of disks attached to a few
%x specialized I/O nodes in the massively parallel processor
%x (MPP). This arrangement requires programmers to give the file
%x system many hints on how their data is to be laid out on disk if
%x they want to achieve good performance. Additionally, the custom
%x interface makes massively parallel file systems hard for
%x programmers to use and difficult to seamlessly integrate into an
%x environment with workstations and tertiary storage.
%x \par
%x The RAMA file system addresses these problems by providing a
%x massively parallel file system that does not need user hints to
%x provide good performance. RAMA takes advantage of the recent
%x decrease in physical disk size by assuming that each processor in
%x an MPP has one or more disks attached to it. Hashing is then used
%x to pseudo-randomly distribute data to all of these disks, insuring
%x high bandwidth regardless of access pattern. Since MPP programs
%x often have many nodes accessing a single file in parallel, the file
%x system must allow access to different parts of the file without
%x relying on a particular node. In RAMA, a file request involves only
%x two nodes -- the node making the request and the node on whose disk
%x the data is stored Thus, RAMA scales well to hundreds of
%x processors. Since RAMA needs no layout hints from applications, it
%x fits well into systems where users cannot (or will not) provide
%x such hints. Fortunately, this flexibility does not cause a large
%x loss of performance. RAMA's simulated performance is within 10-15%
%x of the optimum performance of a similarly-sized striped file
%x system, and is a factor of 4 or more better than a striped file
%x system with poorly laid out data.

%z InProceedings
%K Wakeman95
%A Ian Wakeman
%A Atanu Ghosh
%A Jon Crowcroft
%y CSDEPT., Univ. College London, Gower Street, London WC1E 6BT
%A Van Jacobson
%A Sally Floyd
%y Lawrence Berkeley Laboratory, One Cyclotron Road, Berkeley, CA 94720
%T Implementing real time packet forwarding policies using streams
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 71 82
%x This paper describes an implementation of the class based queuing
%x (CBQ) mechanisms proposed by Sally Floyd and Van Jacobson [1] [2]
%x to provide real time policies for packet forwarding. CBQ allows the
%x traffic flows sharing a data link to be guaranteed a share of the
%x bandwidth when the link is congested, yet allows flexible sharing
%x of the unused bandwidth when the link is unloaded. In addition, CBQ
%x provides mechanisms which give flows requiring low delay priority
%x over other flows. In this way, links can be shared by multiple
%x flows yet still meet the policy and Quality of Service (QoS)
%x requirements of the flows.
%x \par
%x We present a brief description of the implementation and some
%x preliminary performance measurements. The problems of packet
%x classification are addressed in a flexible and extensible, yet
%x efficient manner, and whilst the Streams implementation cannot cope
%x with very high speed interfaces, it can cope with the serial link
%x speeds that are likely to be loaded.

%z InProceedings
%K Schiller95
%A Jeffrey I. Schiller
%A Derek Atkins
%y MIT.
%T Scaling the web of trust: combining Kerberos and PGP to provide large scale authentication
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 83 94
%x Internet Security has become more important recently as the
%x Internet grows exponentially and security breaches become more
%x publicized. An important area of concern for many Internet users is
%x the privacy and integrity of their electronic files and
%x messages. Phil Zimmermann's Pretty Good Privacy PGP) provides a
%x general purpose utility for file and message protection. However
%x PGP requires that communicating users be ``introduced'' to each
%x other. This paper describes a scheme that permits an enterprise
%x using Kerberos to create an automated introducer called the PGP Key
%x Signer Service. Using this service people in the enterprise who
%x have no common acquaintances to act as introducers can be
%x introduced through the Key Signer.

%z InProceedings
%K Kumar95
%A Puneet Kumar
%A M. Satyanarayanan
%y CMU.
%T Flexible and safe resolution of file conflicts
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 95 106
%x In this paper we describe the support provided by the Coda File
%x System for transparent resolution of conflicts arising from
%x concurrent updates to a file in different network partitions. Such
%x partitions often occur in mobile computing environments. Coda
%x provides a framework for invoking customized pieces of code called
%x application-specific resolvers (ASRS) that encapsulate the
%x knowledge needed for file resolution. If resolution succeeds, the
%x user notices nothing more than a slight performance delay. Only if
%x resolution fails does the user have to resort to manual repair. Our
%x design combines a rule-based approach to ASR selection with
%x transactional encapsulation of ASR execution. This paper shows how
%x such an approach leads to flexible and efficient file resolution
%x without loss of security or robustness.

%z InProceedings
%K Dilley95
%A John Dilley
%y HPL.
%T OODCE: a C++ framework for the OSF Distributed Computing Environment
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 107 118
%x This paper presents a method for developing object-oriented
%x distributed applications using the C++ and DCE technologies. The
%x core of this package is a DCE lDL-to-C++ compiler and a set of C++
%x classes providing easy access to DCE functionality.
%x \par
%x Using this approach we were able to develop more object-oriented
%x distributed applications, and saw a significant decrease in
%x application code size. This contributed to an increase in developer
%x productivity and code maintainability.

%z InProceedings
%K Stevenson95
%A J. Mark Stevenson
%y School of Computer Science, CMU.
%A Daniel P. Julin
%y ISIS Distributed Systems
%T Mach-US: UNIX on generic OS object servers
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 119 130
%x This paper examines the Mach-US operating system, its unique
%x architecture, and the lessons demonstrated through its
%x implementation.
%x \par
%x Mach-US is an object-oriented multi-server OS which runs on the
%x Mach3.0 kernel. Mach-US has a set of separate servers supplying
%x orthogonal OS services and a library which is loaded into each user
%x process. This library uses the services to generate the semantics
%x of the Mach2.5/4.3BSD application programmers interface (API). This
%x architecture makes Mach-US a flexible research platform and a
%x powerful tool for developing and examining various OS service
%x options.
%x \par
%x  We will briefly describe Mach-US, the motivations for its design
%x choices, and its demonstrated strengths and weaknesses. We will
%x then discuss the insights that we've acquired in the areas of
%x multiserver architecture, OS remote method invocation, Object
%x Oriented technology for OS implementation, API independent OS
%x services, UNIX API re-implementation, and smart user-space API
%x emulation libraries.

%z InProceedings
%K Waldo95
%A Jim Waldo
%A Ann Wollrath
%A Geoff Wyant
%A Samuel C. Kendall
%y Sun Microsystems Laboratories
%T Events in an RPC based distributed system
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 131 142
%x We show how to build a distributed system allowing objects to
%x register interest in and receive notifications of events in other
%x objects. The system is built on top of a pair of interfaces that
%x are interesting only in their extreme simplicity. We then present a
%x simple and efficient implementation of these interfaces.
%x \par
%x We then show how more complex functionality can be introduced to
%x the system by adding third-party services. These services can be
%x added without t changing the simple interfaces, and without
%x changing the objects in the system that do not need the
%x functionality of those services.
%x \par
%x Finally, we note a number of open issues that remain, and attempt
%x to draw some conclusions based on the work.

%z InProceedings
%K Talbot95
%A Jacques Talbot
%y Bull
%T Turning the AIX operating system into an MP-capable OS
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 143 153
%x This paper describes those MP features that Bull and IBM together
%x introduced into the AIX operating system to support the Symmetric
%x Multiprocessor machine marketed by Bull under the Escala name and
%x by IBM under the RS/6000 Models G30, J30 and R30 names. The PowerPC
%x architecture and the AIX operating system present some specific
%x challenges. We present the major problems encountered and how they
%x were solved

%z InProceedings
%K Kawaguchi95
%A Atsuo Kawaguchi
%A Shingo Nishioka
%A Hiroshi Motoda
%y Advanced Research laboratory, Hitachi, Ltd.
%T A flash-memory based file system
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 155 164
%x  A flash memory device driver that supports a conventional UNIX
%x file system transparently was designed. To avoid the limitations
%x due to flash memory's restricted number of write cycles and its
%x inability to be overwritten, this driver writes data to the flash
%x memory system sequentially as a Log-structured File System (LFS)
%x does and uses a cleaner to collect valid data blocks and reclaim
%x invalid ones by erasing the corresponding flash memory
%x regions. Measurements showed that the overhead of the cleaner has
%x little effect on the performance of the prototype when utilization
%x is low but that the effect becomes critical as the utilization gets
%x higher, reducing the random write throughput from 222 Kbytes/s at
%x 30% utilization to 40 Kbytes/s at 90% utilization. The performance
%x of the prototype in the Andrew Benchmark test is roughly equivalent
%x to that of the 4.4BSD Pageable Memory based File System (MFS).

%z InProceedings
%K Berman95
%A Andrew Berman
%A Virgil Bourassa
%A Erik Selberg
%y DEPTCSE., Univ. of Washington
%T TRON: process-specific file protection for the UNIX operating system
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 165 175
%x The file protection mechanism provided in UNIX is insufficient for
%x current computing environments. While the UNIX file protection
%x system attempts to protect users from attacks by other users, it
%x does not directly address the agents of destruction -- executing
%x processes. As computing environments become more interconnected and
%x interdependent, there is increasing pressure and opportunity for
%x users to acquire and test non-secure, and possibly malicious,
%x software.
%x \par
%x We introduce TRON, a process-level discretionary access control
%x system for UNIX. TRON allows users to specify capabilities for a
%x process' access to individual files, directories, and directory
%x trees. These capabilities are enforced by system call wrappers
%x compiled into the operating system kernel. No privileged system
%x calls, special files, system administrator intervention, or changes
%x to the file system are required. Existing UNIX programs can be run
%x without recompilation under TRON-enhanced UNIX. Thus, TRON improves
%x UNIX security while maintaining current standards of flexibility
%x and openness.

%z InProceedings
%K Yan95
%A Tak W. Yan
%A Hector Garcia-Molina
%y DEPTCS., Stanford Univ.
%T SIFT - a tool for wide-area information dissemination
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 177 186
%x The dissemination model is becoming increasingly important in
%x wide-area information system. In this model, the user subscribes to
%x an information dissemination service by submitting profiles that
%x describe his interests. He then passively receives new, filtered
%x information. The Stanford Information Filtering Tool (SIFT) is a
%x tool to help provide such service. It supports full-text filtering
%x using well-known information retrieval models. The SIFT filtering
%x engine implements novel indexing techniques, capable of processing
%x large volumes of information against a large number of profiles. It
%x runs on several major Unix platforms and is freely available to the
%x public. In this paper we present SIFT's approach to user interest
%x modeling and user-server communication. We demonstrate the
%x processing capability of SIFT by describing a running server that
%x disseminates USENET News. We present an empirical study of SIFT's
%x performance, examining its main memory requirement and ability to
%x scale with information volume and user population .

%z InProceedings
%K Mogul95
%A Jeffrey C. Mogul
%A Joel F. Bartlett
%A Robert N. Mayo
%A Amitabh Srivastava
%y DECWRL.
%T Performance implications of multiple pointer sizes
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 187 200
%x Many users need 64-bit architectures: 32-bit systems cannot support
%x the largest applications, and 64bit systems perform better for some
%x applications. However, performance on some other applications can
%x suffer from the use of large pointers; large pointers can also
%x constrain feasible problem size. Such applications are best served
%x by a 64-bit machine that supports the use of both 32-bit and 64-bit
%x pointer variables.
%x \par
%x This paper analyzes several programs and programming techniques to
%x understand the performance implications of different pointer
%x sizes. Many (but not all) programs show small but definite
%x performance consequences, primarily due to cache and paging
%x effects.

%z InProceedings
%K Golding95
%A Richard Golding
%A Peter Bosch
%A Carl Staelin
%A Tim Sullivan
%A John Wilkes
%y HPL.
%T Idleness is not sloth
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 201 212
%x Many people have observed that computer systems spend much of their
%x time idle, and various schemes have been proposed to use this idle
%x time productively. The commonest approach is to off-load activity
%x from busy periods to less-busy ones in order to improve system
%x responsiveness. In addition, speculative work can be performed in
%x idle periods in the hopes that it will be needed later at times of
%x higher utilization, or non-renewable resource like battery power
%x can be conserved by disabling unused resources.
%x \par
%x We found opportunities to exploit idle time in our work on storage
%x systems, and after a few attempts to tackle specific instances of
%x it in ad hoc ways, began to investigate general mechanisms that
%x could be applied to this problem. Our results include a taxonomy of
%x idle time detection algorithms, metrics for evaluating them, and an
%x evaluation of a number of idleness predictors that we generated
%x from our taxonomy.

%z InProceedings
%K Plank95
%A James S. Plank
%A Micah Beck
%A Gerry Kingsley
%y Univ. of Tennessee
%A Kai Li
%y Princeton Univ.
%T Libckpt: transparent checkpointing under Unix
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 213 224
%x Checkpointing is a simple technique for rollback recovery: the
%x state of an executing program is periodically saved to a disk file
%x from which it can be recovered after a failure. While recent
%x research has developed a collection of powerful techniques for
%x minimizing the overhead of writing checkpoint files, checkpointing
%x remains unavailable to most application developers. In this paper
%x we describe libckpt, a portable checkpointing tool for Unix that
%x implements all applicable performance optimizations which are
%x reported in the literature. While libckpt can be used in a mode
%x which is almost totally transparent to the programmer, it also
%x supports the incorporation of user directives into the creation of
%x checkpoints. This user-directed checkpointing is an innovation
%x which is unique to our work.

%z InProceedings
%K Ho95
%A W. Wilson Ho
%A Wei-Chau Chang
%A Lilian H. Leung
%y Silicon Graphics, Inc.
%T Optimizing the performance of dynamically-linked programs
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 225 233
%x Dynamically-linked programs in general do not perform as well as
%x statically-linked programs. This paper identifies three main areas
%x that account for the performance loss. First, symbols are
%x referenced indirectly and thus extra instructions are
%x required. Second, the overhead in run-time symbol resolution is
%x significant. Third, poor locality of functions in shared libraries
%x and data structures maintained by the run-time linker may result in
%x poor memory utilization. This paper presents new optimization
%x techniques we developed that address these three areas and
%x significantly improve the performance of dynamically-linked
%x programs. Also, we provide measurements of the performance
%x improvement achieved. Most importantly, we show that all desirable
%x features of shared libraries can be achieved without sacrificing
%x performance.

%z InProceedings
%K Arnow95
%A David M Arnow
%y Brooklyn College, CUNY
%T DP: A library for building portable, reliable distributed applications
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 235 247
%x DP is a library of process management and communication tools for
%x writing portable, reliable distributed applications. It provides
%x support for a flexible set of message operations as well as process
%x creation and management. It has been successfully used in
%x developing distributed Monte Carlo, disjunctive programming and
%x integer goal programming codes.It differs from PVM and similar
%x libraries in its support for lightweight, unreliable messages, as
%x well as asynchronous delivery of interrupt-generating messages. In
%x addition, DP supports the development of long-running distributed
%x applications tolerant to the failure or loss of a subset of its
%x processors.

%z InProceedings
%K Seltzer95
%A Margo Seltzer
%A Keith A. Smith
%y Harvard Univ.
%A Hari Balakrishnan
%A Jacqueline Chang
%A Sara McMains
%A Venkata Padmanabhan
%y UCB.
%T File system logging versus clustering: a performance comparison
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 249 264
%x The Log-structured File System (LFS), introduced in 1991 [8], has
%x received much attention for its potential order-of-magnitude
%x improvement in file system performance. Early research results [9]
%x showed that small file performance could scale with processor speed
%x and that cleaning costs could be kept low, allowing LFS to write at
%x an effective bandwidth of 62 to 83% of the maximum. Later work
%x showed that the presence of synchronous disk operations could
%x degrade performance by as much as 62% and that cleaning overhead
%x could become prohibitive in transaction processing workloads,
%x reducing performance by as much as 40% [10]. The same work showed
%x that the addition of clustered reads and writes in the Berkeley
%x Fast File System [6] (FFS) made it competitive with LFS in
%x large-file handling and software development environments as
%x approximated by the Andrew benchmark [4]
%x \par
%x These seemingly inconsistent results have caused confusion in the
%x file system research community. This paper presents a detailed
%x performance comparison of the 4.4BSD Log-structured File System and
%x the 4.4BSD Fast File System. Ignoring cleaner overhead, our results
%x show that the order-of-magnitude improvement in performance claimed
%x for LFS applies only to meta-data intensive activities,
%x specifically the creation of files one-kilobyte or less and
%x deletion of files 64 kilobytes or less.
%x \par
%x For small files, both systems provide comparable read performance,
%x but LFS offers superior performance on writes For large files (one
%x megabyte and larger), the performance of the two file systems is
%x comparable. When FFS is tuned for writing, its large-file write
%x performance is approximately 15% better than LFS, but its read
%x performance is 25% worse. When FFS is optimized for reading, its
%x large-file read and write performance is comparable to LFS.
%x \par
%x Both LFS and FFS can suffer performance degradation, due to
%x cleaning and disk fragmentation respectively. We find that active
%x FFS file systems function at approximately 85-95% of their maximum
%x performance after two to three years. We examine LFS cleaner
%x performance in a transaction processing environment and find that
%x cleaner overhead reduces LFS performance by more than 33% when the
%x disk is 50% full.

%z InProceedings
%K Vahalia95
%A Uresh Vahalia
%y EMC Corporation
%A Cary G. Gray
%y Abilene Christian Univ.
%A Dennis Ting
%y EMC Corporation
%T Metadata logging in an NFS server
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 265 276
%x Over the last few years, there have been several efforts to use
%x logging to improve performance, reliability, and recovery times of
%x file systems. The two major techniques are metadata logging, where
%x the log records metadata changes and is a supplement to the on-disk
%x file system, and log structured file systems, whose log is their
%x only ondisk representation. When the file system is mainly or
%x wholly accessed through the Network File System (NFS) protocol, it
%x adds new considerations to the suitability of the logging technique
%x NFS requires that all operations be updated to stable storage
%x before returning. As a result, file system implementations that
%x were effective for local access may perform poorly on an NFS
%x server. This paper analyzes the issues regarding the use of logging
%x on an NFS server, and describes an implementation of a BSD Fast
%x File System (FFS) with metadata logging that performs effectively
%x for a dedicated NFS server.

%z InProceedings
%K Blackwell95
%A Trevor Blackwell
%A Jeffrey Harris
%A Margo Seltzer
%y Harvard Univ.
%T Heuristic cleaning algorithms in log-structured file systems
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 277 288
%x Research results show that while Log Structured File Systems (LFS)
%x offer the potential for dramatically improved file system
%x performance, the cleaner can seriously degrade performance, by as
%x much as 40% in transaction processing workloads [9]. Our goal is to
%x examine trace data from live file systems and use those to derive
%x simple heuristics that will permit the cleaner to run without
%x interfering with normal file access Our results show that trivial
%x heuristics perform very well, allowing 97% of all cleaning on the
%x most heavily loaded system we studied to be done in the background.

%z InProceedings
%K Ramsey95
%A Norman Ramsey
%y Bell Communications Research
%A Mary F. Fernandez
%y DEPTCS., Princeton Univ.
%T The New Jersey Machine-Code Toolkit
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 289 302
%x The New Jersey Machine-Code Toolkit helps programmers write
%x applications that process machine code. Applications that use the
%x toolkit are written at an assembly-language level of abstraction,
%x but they recognize and emit binary. Guided by a short
%x instructionset specification, the toolkit generates all the
%x bitmanipulating code.
%x \par
%x The toolkit's specification language uses four concepts: fields and
%x tokens describe parts of instructions, patterns describe binary
%x encodings of instructions or groups of instructions, and
%x constructors map between the assembly-language and binary
%x levels. These concepts are suitable for describing both CISC and
%x RISC machines; we have written specifications for the MIPS R3000,
%x SPARC, and Intel 486 instruction sets.
%x \par
%x We have used the toolkit to help write two applications: a debugger
%x and a linker. The toolkit generates efficient code; for example,
%x the linker emits binary up to 15% faster than it emits assembly
%x language, making it 1.7-2 times faster to produce an a.out directly
%x than by using the assembler.

%z InProceedings
%K Eustace95
%A Alan Eustace
%A Amitabh Srivastava
%y DECWRL.
%T ATOM: a flexible interface for building high performance program analysis tools
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 303 314
%x Code instrumentation is a powerful mechanism for understanding
%x program behavior. Unfortunately, code instrumentation is extremely
%x difficult, and therefore has been mostly relegated to building
%x special purpose tools for use on standard industry benchmark
%x suites.
%x \par
%x ATOM (Analysis Tools with OM) provides a very flexible and
%x efficient code instrumentation interface that allows powerful, high
%x performance program analysis tools to be built with very little
%x effort. This paper illustrates this flexibility by building five
%x complete tools that span the interests of application programmers,
%x computer architects, and compiler writers.
%x \par
%x This flexibility does not come at the expense of
%x performance. Because AIOM uses procedure calls as the interface
%x between the application and the analysis routines, the performance
%x of each tool is similar to or greatly exceeds the best known
%x hand-crafted implementations.

%z InProceedings
%K Graham95
%A Susan L. Graham
%A Steven Lucco
%A Robert Wahbe
%y Comp. Sci. Division, UCB.
%T Adaptable binary programs
%C Conf. Proc. of the USENIX 1995 Tech. Conf. on UNIX and Advanced Compt. Sys.
%c New Orleans
%p Usenix Assoc.
%I ISBN 1--880446--67--7
%D 16--20 Jan. 1995
%P 315 325
%x To accurately and comprehensively monitor a program's behavior,
%x many performance measurement tools transform the program's
%x executable representation or binary. By instrumenting binary
%x programs to monitor program events, tools can precisely analyze
%x compiler optimization effectiveness, memory system performance,
%x pipeline interlocking, and other dynamic program characteristics
%x that are fully exposed only at this level. Binary transformation
%x has also been used to support software-enforced fault isolation,
%x debugging, machine re-targeting and machine-dependent optimization
%x \par
%x At present, binary transformation applications face a difficult
%x trade-off. Previous approaches to implementing robust
%x transformations result in significant disk space and run-time
%x overhead. To improve efficiency, some current systems sacrifice
%x robustness, relying on heuristic assumptions about the program and
%x recognition of compiler dependent code generation idioms In this
%x paper we begin by investigating the run-time and disk space
%x overhead of transformation strategies that do not require
%x assumptions about the program's control flow or register usage. We
%x then detail simple information about the binary program that can
%x significantly reduce this overhead. For each type of information,
%x we show how it enables a corresponding type of binary
%x transformation. We call binary programs that contain such enabling
%x information adaptable binaries. Because adaptable binary
%x information is simple, any compiler can generate it. Despite its
%x simplicity, adaptable binary information has the necessary and
%x sufficient expressive power to support a rich set of binary
%x transformations.
%x \par
%x We evaluated a prototype adaptable binary transformation system
%x under the Ultrix 4.2 operating system and the MIPS processor
%x architecture Using the C SPEC92 benchmarks, we assessed adaptable
%x binaries in two ways. First, we demonstrated that the information
%x necessary to build adaptable binaries can be compactly recorded,
%x increasing space overhead by only 9NO for the SPEC92
%x benchmarks. Second, we measured the run-time overhead of previous
%x approaches to implementing robust binary transformations, and
%x showed that adaptable binaries enable a significant reduction of
%x this overhead.

You are viewing proxied material from infinitelyremote.com. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.