Newsgroups: comp.org.usenix
From:
[email protected] (Jim Duncan)
Subject: Cincinnati Abstracts
Nntp-Posting-Host: leibniz.math.psu.edu
Organization: Penn State University Department of Mathematics
Date: Wed, 26 May 1993 04:00:51 GMT
This post contains the abstracts for all but three of the papers to be
presented at the upcoming USENIX conference in Cincinnati.
This is an experiment. It's hoped that conference-goers will be able to use
this posting to plan which session to attend in advance, instead of skimming
the proceedings as soon as they've arrived. For those who are still
wondering whether to attend or not, it's hoped that the abstracts will sway
you into coming. Regardless of how you use it, please let me know if you
like it so USENIX can decide whether or not to make it a regular thing.
I plan to post this again about a week before the conference.
Jim
--
Adaptive Block Rearrangement Under UNIX
Sedat Aky"urek & Kenneth Salem
Department of Computer Science
University of Maryland, College Park, MD 20742
Abstract
An adaptive UNIX disk device driver is described. The driver
copies frequently-referenced blocks from their original locations
to reserved space near the center of the disk to reduce seek
times. Reference frequencies need not be known in advance. In-
stead, they are estimated by monitoring the stream of arriving
requests. Measurements show that the adaptive driver reduces
seek times by more than half, and improves response times signi-
ficantly.
--
The Autofs Automounter
Brent Callaghan, Satinder Singh
SunSoft Inc.
Abstract
Prior to the introduction of the automounter in 1987, NFS mounts were
administered separately on each workstation. The automounter has
provided administrators with a tool to construct a filesystem namespace
that can be shared across an organization. While the automounter is
widely used, its success has been tempered by problems inherent in its
implementation. This paper describes a new implementation of the
automounter based on a new filesystem. This new automounter not only
fixes the problems, but provides some interesting opportunities for
future development.
--
Optimizing Unix Resource Scheduling for User Interaction
Steve Evans, Kevin Clarke, Dave Singleton, Bart Smaalders
SunSoft Inc.
Abstract
Techniques for improving system responsiveness for
interactive end users of Unix workstations are
explored. After a discussion of the current state of
resource scheduling, a model is presented in which
dynamic input from the human user is combined with data
from user interaction software to supply a centralized
manager with the information necessary to determine
what processes are involved with interacting with the
user at any given moment. This service then
communicates this process set information to the
kernel, which uses it to manage memory and CPU resource
allocation on the behalf of the user. Experience with
a prototype of this environment is reported. An
argument for an interactive scheduling class is made,
along with other infrastructure changes needed to take
advantage of it.
--
The Shell as a Service
Glenn Fowler
[email protected]
AT&T Bell Laboratories
Murray Hill, New Jersey 07974
Abstract
This paper explores the design history of the nmake shell
coprocess. Originally a special purpose uniprocessor
executor, the coshell has evolved into a general purpose
service that automatically executes shell actions on lightly
loaded hosts in a local network. A major thrust of this
work has been ease of use. The only privilege required for
installation, administration or use is rsh access to the
local hosts.
nmake and GNU-make users can take advantage of network
execution with no makefile modifications. Shell level
access is similar to but more efficient than rsh and allows
host expression matching to replace the explicit host name
argument. Also provided is a C programming library
interface with five primitive operations that follow the
fork-exec-wait process model.
Beside the speedups attained by parallelizing computations
in a homogeneous network, coshell also supports
heterogeneous configurations. This presents novel solutions
to traditional cross-compilation problems. It also allows
the user to view a new network host as a compute engine
rather than yet another architecture on which to port the
home environment and tools.
coshell runs on most S5R4 and BSD UNIX* operating system
variants.
--
Call Path Profiling of Monotonic Program Resources in UNIX
by
Robert J. Hall
Aaron J. Goldberg
AT&T Bell Laboratories
600 Mountain Ave.
Murray Hill, NJ 07974
Abstract
Practical performance improvement of a complex program
must be guided by empirical measurements of its resource usage.
Essentially, the programmer wants to know where in the source code
the program is inefficient and why this is so.
The process interface of UNIX System V (proc(4))
provides access to the raw data (e.g. time, faults, traps, and system
calls) necessary to answering the why question,
but gives no guidance in answering the where question.
This paper describes a novel approach to the latter,
Call Path Profiling, which is both more informative
and more closely tied to the process of program optimization than
either trace-based or prof/gprof-like approaches.
In addition, by viewing consumption of a resource as the ticking of a
clock, we generalize the interval-based sampling approach of
time profilers to arbitrary monotonic resources.
The approach is embodied in several prototypes, including CPPROF which
operates under System V.
--
Computer System Performance Problem Detection
Using Time Series Models
Peter Hoogenboom and Jay Lepreau
University of Utah
Abstract
Computer systems require monitoring to detect performance anomalies
such as runaway processes, but problem detection and diagnosis is a
complex task requiring skilled attention. Although human attention
was never ideal for this task, as networks of computers grow larger
and their interactions more complex, it falls far short. Existing
computer-aided management systems require the administrator manually
to specify fixed "trouble" thresholds. In this paper we report on an
expert system that automatically sets thresholds, and detects and
diagnoses performance problems on a network of Unix computers. Key
to the success and scalability of this system are the time series
models we developed to model the variations in workload on each host.
Analysis of the load average records of 50 machines yielded models
which show, for workstations with simulated problem injection, false
positive and negative rates of less than 1%. The server machines most
difficult to model still gave average false positive/negative rates
of only 6%/32%. Observed values exceeding the expected range for a
particular host cause the expert system to focus on that machine.
There it applies tools with finer resolution and more discrimination,
including per-command profiles gleaned from process accounting records.
It makes one of 18 specific diagnoses and notifies the administrator,
and optionally the user [a].
--
The Ferret Document Browser
Howard P. Katseff
Thomas B. London
AT&T Bell Laboratories
Holmdel, NJ 07733
Abstract
The Ferret Document Browser is a vehicle for exploring the
design and use of document storage and retrieval systems.
Its distributed, modular structure allows independent infor-
mation providers to control their data, yet make use of a
common access and billing control facility. Document images
are distributed via a nationwide AT&T corporate internet
which consists mainly of Ethernet networks interconnected by
leased data circuits. The relatively low bandwidth of this
networks is dealt with by compressing the documents for
transmission, and by decompressing pages as requested on the
workstation. A page image can be decompressed and displayed
in less that a half second.
A broadband version of the system makes use of the BBFS
broadband file server, the HPC interconnect, the LuckyNet
broadband network and the Liaison network multimedia work-
station. This system allows document browsing at rates up
to 15 page images per second.
--
Integrating Handwriting Recognition into Unix
James Kempf
Nomadic Systems Group,
Sun Microsystems Computer Corp.
2550 Garcia Ave., Mail Stop MTV17-08
Mountain View, CA, 94043
Abstract
Many new portable computers are substituting an electronic
stylus, or pen, for the mouse. While the pen can serve as a sim-
ple replacement for the mouse, it also provides an enhanced draw-
ing capability. This capability opens up the potential for new
modes of user interaction, one of which is text input through
handwriting instead of keyboard entry. In this paper, the in-
tegration of handwriting recognition into the Unix operation sys-
tem is discussed. We begin with an examination of the current
state of the art in recognition algorithms and how handwriting
recognition can enhance a user interface. A standard applica-
tion program interface for handwriting recognition engines (HRE
API) is then presented. The HRE API is distinguished from exist-
ing PC operating system API's in that it is specifically designed
for multiple handwriting recognition engines of differing techno-
logies, rather than a single, vendor-specific engine, and it
shares a relatively narrow surface area with the window system.
The latter characteristic allows it to be used with existing win-
dow systems, such as X, but does not hinder migration to other
window systems should they become available. The API has been
implemented with a public domain recognition engine and is
currently being circulated among vendors of handwriting recogni-
tion engines for comment. Finally, the paper concludes with a
discussion of where handwriting recognition belongs in the
current X window system architecture, and what would be needed to
make handwriting an equal partner with typed keyboard input for
text entry.
--
Design And Implementation of a Multimedia Protocol Suite
in a BSD Unix Kernel
Lakshman K, Giri Kuthethoor, Raj Yavatkar Department of Computer
Science University of Kentucky, Lexington, KY 40506
Abstract
Development of distributed multimedia applications
requires support for coordination and temporal/causal
synchronization of traffic over related streams. Our current
research involves investigation of appropriate OS and
communication abstractions to support such applications. Towards
this goal, we have designed and implemented MCP, a suite of
transport and session layer protocols, in the framework of a
standard BSD Unix networking platform. MCP contains two new
abstractions. First, MCP contains a token-based mechanism for
coordination of traffic over a multipoint connection. Second,
MCP includes an abstraction called a multi-flow conversation
that enforces both temporal and causal synchronization among
related data streams. This paper discusses Unix kernel
implementation of MCP and describes our experience in using MCP.
--
sfs: A Parallel File System for the CM-5*
Susan J. LoVerso Marshall Isman Andy Nanopoulos
William Nesheim Ewan D. Milne Richard Wheeler
CM-5 Operating System Group
Thinking Machines Corporation
245 First Street
Cambridge, MA 02142-1264
Abstract
This paper describes the creation of a Unix-compatible
file system with highly scalable performance and size.
The file system is on the CM-5 backed by a scalable array
of disks. Using the Unix file system (UFS) from the SunOS 4.1.2 ker-
nel as a base and modifying it to support Connection Machine
(CM) operations, we have created a new file system, the scalable
file system, or sfs. We discuss the CM operations we support,
such as parallel reads and writes to the processing nodes of the
Connection Machine, the use of NFS to support many partitions
of processing nodes on the CM, support for very large file
sizes (64-bit) and support for odd numbers of disk drives. The
tradeoffs and decisions made during the course of this project
as well as performance data for varying numbers of disk
drives are provided.
--
The Restore-o-Mounter
The File Motel Revisited
Joe Moran
Bob Lyon
Legato Systems, Incorporated
Abstract
We present a scheme for referencing and accessing saved
(( footnote 1: We use the word "save" to denote the super set of
"backup" and "archive"; save is also easier to conjugate than backup.))
files in a manner that
is transparent to UNIX" applications. The scheme requires no kernel
modifications. Instead, it uses a "mounted" process that allows users
to change directories to the past and browse their saved files with
their favorite utilities. The mounted process acts as a protocol
gateway between NFS and a commercially available network backup
product. Time travel is supported; users may change directories to any
moment in the past. Any saved version (not just the most recent
version) of any file can be viewed or recovered, even if the file has
since been deleted.
Using this transparent method of retrieving saved files by naming their
location in the past, a poor man's file migration scheme can be
implemented by substituting a symbolic link to a saved location for a
file. Once a file is referenced, the symbolic link can be replaced
with its original file. This migration scheme requires no kernel
modifications yet remains transparent to UNIX applications and users.
--
High Performance Dynamic Linking Through Caching
Michael N. Nelson Graham Hamilton
Sun Microsystems Laboratories, Inc.
Mountain View, CA 94043 USA
Abstract
The Spring Operating System provides high performance dynamic
linking of program images. Spring uses caching of both fixed-up
program images and partially fixed-up shared libraries to make
dynamic linking of program images efficient, to reduce the need
for PIC (position-independent code), and to improve page sharing
between different program images running the same libraries. The
result is that with program image caching, dynamically-linked
programs have a start-up cost similar to statically-linked pro-
grams regardless of the number of unresolved symbols in
dynamically-linked program images and shared libraries. In addi-
tion, with library and program image caching, we have reduced the
need for PIC and have increased page sharing.
--
Design and Implementation of a Simulation Library using
Lightweight Processes
Janche Sang
Ke-hsiung Chung
Vernon Rego
Department of Computer Sciences
Purdue University
West Lafayette, IN 47907
Abstract
The \Si/ lightweight-process based system for simulating
process interactions is an enhancement to the C programming
language in the form of library primitives with sets of predefined
data structures. The \Si/ system encapsulates an existing
lightweight-process library
to provide a discrete-event simulation environment supporting the
process view. It was developed as a research testbed for investigating
methods which support simulations efficiently. Easy extensions and
modifications to the \Si/ system are a major design objective, accomplished
through modularity and layering. This paper describes the system,
our experiences with its implementation, and its applicability
to simulation modeling. We report on performance measurements of different
implementations of the simulation scheduler, and of different algorithms for simulating service disciplines.
--
``Stacking'' Vnodes: A Progress Report
Glenn C. Skinner and Thomas K. Wong
SunSoft Inc.
2550 Garcia Avenue
Mountain View, Ca 94043
ABSTRACT
People are dissatisfied with the file system ser-
vices that come with their UNIX systems. They want to
add new and better features. At present they have two
choices: express their service as a user-level NFS
server, or use the vnode/VFS interface to build at
least part of it into the kernel. Although the
vnode/VFS interface has been remarkably successful as
a kernel structuring concept, it has failed to provide
source portability between UNIX versions or even
binary compatibility between releases of the same UNIX
version. It has been obvious for some time that a
redesign of the vnode/VFS interface that allowed file
systems to be shipped as binary kernel modules that
survive from release to release is needed. We
describe a prototype kernel with a vnode/VFS interface
that would allow this. It is based on earlier work on
``stacking'' vnodes at Sun and at UCLA, but it
replaces the stacking concept by a more strictly
object-oriented concept of interposition.
--
AudioFile: A Network-Transparent System
for Distributed Audio Applications
Thomas M. Levergood, Andrew C. Payne,
James Gettys, G. Winfeld Treese, and Lawrence C. Stewart
Digital Equipment Corporation
Cambridge Research Lab
Abstract
AudioFile is a portable, device-independent, network-transparent
system for computer audio systems. Similar to the X Window Sys-
tem, it provides an abstract audio device interface with a simple
network protocol to support a variety of audio hardware and mul-
tiple simultaneous clients. AudioFile emphasizes client handling
of audio data and permits exact control of timing. This paper
describes our approach to digital audio, the AudioFile protocol,
the client library, the audio server, and some example client ap-
plications. It also discusses the performance of the system and
our experience using standard networking protocols for audio. A
source code distribution is available by anonymous FTP.
--
X Through the Firewall, and Other Application Relays
G. Winfeld Treese
MIT Laboratory for Computer Science and Digital Equipment Corporation
Alec Wolman
University of Washington and Digital Equipment Corporation
Abstract
Organizations often impose an administrative security policy when they
co nnect to other organizations on a public network such as the
Internet. Many applications have their own notions of security,
or they simply rely on the security of the underlying protocols. Using
the X Window System as a case study, we describe some techniques for
building application-speci#c #relays# that allow the use of
applications across organizational boundaries. In particular, we focus
on analyzing administrative and application-speci#c security policies
to construct solutions that satisfy the security requirements
while providing the necessary functions of the applications.
--
LADDIS: The Next Generation In NFS File Server Benchmarking
Mark Wittle, Data General Corporation
Bruce E. Keith, Digital Equipment Corporation
Abstract
The ability to compare the performance of various NFS(1) file
server configurations from several vendors is critically important to a
computing facility when selecting an NFS file server. To date,
nhfsstone(2) has been a popular means of characterizing NFS file server
performance. However, several deficiencies have been found in
nhfsstone. The LADDIS NFS file server benchmark has been developed to
resolve nhfsstone's shortcomings and provide new functionality. The
Standard Performance Evaluation Corporation (SPEC(3)) released the
System File Server (SFS) Release 1.0 benchmark suite, which contains
097.LADDIS, as an industry-standard NFS file server benchmark in April
1993. This paper describes the major technical issues involved in
developing the benchmark and the rationale used to establish default
097.LADDIS workload parameter values. Where appropriate, areas for
further research are identified and encouraged.
--
Anonymous RPC:
Low-Latency Protection in a 64-Bit Address Space
Curtis Yarvin, Richard Bukowski, and Thomas Anderson
Computer Science Division, University of California at Berkeley
Abstract:
In this paper, we propose a method of reducing the latency of
cross-domain remote procedure call (RPC). Traditional systems use
separate address spaces to provide memory protection between separate
processes, but even with a highly optimized RPC system, the cost of
switching between address spaces can make cross-domain RPC
prohibitively expensive.
Our approach is to use _anonymity_ instead of hardware page tables for
protection. Logically independent memory segments are placed at random
locations in the same address space and protection domain. With 64-bit
virtual addresses, it is unlikely that a process will be able to
locate any other segment by accidental or malicious memory probes; it
is impossible to corrupt a segment without knowing its location. The
benefit is that a cross-domain RPC need not involve a hardware context
switch. Measurements of our prototype implementation show that a
round-trip null RPC takes only 7.7us on an Intel 486-33.
--
Discovery and Hot Replacement
of Replicated Read-Only File Systems,
with Application to Mobile Computing
Erez Zadok and Dan Duchamp
Computer Science Department
Columbia University
ABSTRACT
We describe a mechanism for replacing files, including open files, of a
read-only file system while the file system remains mounted; the act of
replacement is transparent to the user. Such a "hot replacement"
mechanism can improve fault-tolerance, performance, or both. Our
mechanism monitors, from the client side, the latency of operations
directed at each file system. When latency degrades, the client
automatically seeks a replacement file system that is equivalent
to but hopefully faster than the current file system. The files
in the replacement file system then take the place of those in the
current file system. This work has particular relevance to mobile
computers, which in some cases might move over a wide area. Wide
area movement can be expected to lead to highly variable response
time, and give rise to three sorts of problems: increased latency,
increased failures, and decreased scalability. If a mobile client
moves through regions having partial replicas of common file systems,
then the mobile client can depend on our mechanism to provide
increased fault tolerance and more uniform performance.
--
--
Jim Duncan <
[email protected]> Penn State Math Dept Systems Administrator
"[A computer is] like an Old Testament god, with a lot of rules and no mercy."
Joseph Campbell
