%z Article
%K Banga97
%A Guarav Banga
%A Peter Druschel
%T Measuring the capacity of a web server
%B Proceedings USENIX Symposium on Internet Technologies and Systems
%C Monterey, CA
%D December 1997

%z Article
%K Banga98
%A Guarav Banga
%A Jeffrey C. Mogul
%T Scalable kernel performance for internet servers under realistic loads
%B Proceedings of the 1998 USENIX Annual Technical Conference
%C New Orleans, LA
%D June 1998

%K Bray90
%A Tim Bray
%T Bonnie benchmark
%D 1990
%O http://www.textuality.com/bonnie/

%z Article
%K Brown97
%A Aaron Brown
%A Margo Seltzer
%T Operating system benchmarking in the wake of lmbench: A case study of the performance of NetBSD on the Intel x86 architecture
%B Proceedings of the 1997 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems
%C Seattle, WA
%D June 1997
%P 214-224
%O http://www.eecs.harvard.edu/~vino/perf/hbench/sigmetrics/hbench.html

%z Article
%K Chen93d
%A Peter M. Chen
%A David Patterson
%T Storage performance \- metrics and benchmarks
%J Proceedings of the IEEE
%V 81
%N 8
%D August 1993
%P 1151-1165
%x Discusses metrics and benchmarks used in storage performance evaluation.
%x Describes, reviews, and runs popular I/O benchmarks on three systems.  Also
%x describes two new approaches to storage benchmarks: LADDIS and a Self-Scaling
%x benchmark with predicted performance.
%k I/O, storage, benchmark, workload, self-scaling benchmark, 
%k predicted performance, disk, performance evaluation
%s staelin%cello@hpl.hp.com (Wed Sep 27 16:21:11 PDT 1995)

%z Article
%K Chen94a
%A P. M. Chen
%A D. A. Patterson
%T A new approach to I/O performance evaluation \- self-scaling I/O benchmarks, predicted I/O performance
%D November 1994
%J Transactions on Computer Systems
%V 12
%N 4
%P 308-339
%x Current I/O benchmarks suffer from several chronic problems: they
%x quickly become obsolete; they do not stress the I/O system; and they
%x do not help much in undelsi;anding I/O system performance. We
%x propose a new approach to I/O performance analysis. First, we
%x propose a self-scaling benchmark that dynamically adjusts aspects of
%x its workload according to the performance characteristic of the
%x system being measured. By doing so, the benchmark automatically
%x scales across current and future systems. The evaluation aids in
%x understanding system performance by reporting how performance varies
%x according to each of five workload parameters. Second, we propose
%x predicted performance, a technique for using the results from the
%x self-scaling evaluation to estimate quickly the performance for
%x workloads that have not been measured. We show that this technique
%x yields reasonably accurate performance estimates and argue that this
%x method gives a far more accurate comparative performance evaluation
%x than traditional single-point benchmarks. We apply our new
%x evaluation technique by measuring a SPARCstation 1+ with one SCSI
%x disk, an HP 730 with one SCSI-II disk, a DECstation 5000/200 running
%x the Sprite LFS operating system with a three-disk disk array, a
%x Convex C240 minisupercomputer with a four-disk disk array, and a
%x Solbourne 5E/905 fileserver with a two-disk disk array.
%s toc@hpl.hp.com (Mon Mar 13 10:57:38 1995)
%s wilkes%hplajw@hpl.hp.com (Sun Mar 19 12:38:01 PST 1995)
%s wilkes%cello@hpl.hp.com (Sun Mar 19 12:38:53 PST 1995)

%z Article
%K Fenwick95
%A David M. Fenwick
%A Denis J. Foley
%A William B. Gist
%A Stephen R. VanDoren
%A Danial Wissell
%T The AlphaServer 8000 series: high-end server platform development
%J Digital Technical Journal
%V 7
%N 1
%D August 1995
%P 43-65
%x The AlphaServer 8400 and the AlphaServer 8200 are Digital's newest high-end
%x server products.  Both servers are based on the 300Mhz Alpha 21164 
%x microprocessor and on the AlphaServer 8000-series platform architecture.
%x The AlphaServer 8000 platform development team set aggressive system data
%x bandwidth and memory read latency targets in order to achieve high-performance
%x goals.  The low-latency criterion was factored into design decisions made at
%x each of the seven layers of platform development.  The combination of 
%x industry-leading microprocessor technology and a system platform focused
%x on low latency has resulted in a 12-processor server implementation ---
%x the AlphaServer 8400 --- capable of supercomputer levels of performance.
%k DEC Alpha server, performance, memory latency
%s staelin%cello@hpl.hp.com (Wed Sep 27 17:27:23 PDT 1995)

%z Book
%K Hennessy96
%A John L. Hennessy
%A David A. Patterson
%T Computer Architecture A Quantitative Approach, 2nd Edition
%I Morgan Kaufman
%D 1996

%z Article
%K Howard88
%A J. Howard
%A M. Kazar
%A S. Menees
%A S. Nichols
%A M. Satyanrayanan
%A R. Sidebotham
%A M. West
%T Scale and performance in a distributed system
%J ACM Transactions on Computer Systems
%V 6
%N 1
%D February 1988
%P 51-81
%k Andrew benchmark

%z Book
%K Jain91
%A Raj Jain
%T The Art of Computer Systems Performance Analysis: Techniques for Experimental Design, Measurement, Simulation, and Modeling
%I Wiley-Interscience
%C New York, NY
%D April 1991

%z Article
%K McCalpin95
%A John D. McCalpin
%T Memory bandwidth and machine balance in current high performance computers
%J IEEE Technical Committee on Computer Architecture newsletter
%D December 1995

%z InProceedings
%K McVoy91
%A L. W. McVoy
%A S. R. Kleiman
%T Extent-like Performance from a Unix File System
%B Proceedings USENIX Winter Conference
%C Dallas, TX
%D January 1991
%P 33-43

%z Article
%K McVoy96
%A Larry McVoy
%A Carl Staelin
%T lmbench: Portable tools for performance analysis
%B Proceedings USENIX Winter Conference
%C San Diego, CA
%D January 1996
%P 279-284

%z InProceedings
%K Ousterhout90
%s wilkes%cello@hplabs.hp.com (Fri Jun 29 20:46:08 PDT 1990)
%A John K. Ousterhout
%T Why aren't operating systems getting faster as fast as hardware?
%B Proceedings USENIX Summer Conference
%C Anaheim, CA
%D June 1990
%P 247-256
%x This paper evaluates several hardware pplatforms and operating systems using
%x a set of benchmarks that stress kernel entry/exit, file systems, and
%x other things related to operating systems. The overall conclusion is that
%x operating system performance is not improving at the same rate as the base speed of the
%x underlying hardware. The most obvious ways to remedy this situation
%x are to improve memory bandwidth and reduce operating systems'
%x tendency to wait for disk operations to complete.
%o Typical performance of 10-20 MIPS cpus is only 0.4 times what
%o their raw hardware performance would suggest. HP-UX is
%o particularly bad on the HP 9000/835, at about 0.2x. (Although
%o this measurement discounted a highly-tuned getpid call.)
%k OS performance, RISC machines, HP9000 Series 835 system calls

%z Article
%K Park90a
%A Arvin Park
%A J. C. Becker
%T IOStone: a synthetic file system benchmark
%J Computer Architecture News
%V 18
%N 2
%D June 1990
%P 45-52
%o this benchmark is useless for all modern systems; it fits
%o completely inside the file system buffer cache.  Soon it may even
%o fit inside the processor cache!
%k IOStone, I/O, benchmarks
%s staelin%cello@hpl.hp.com (Wed Sep 27 16:37:26 PDT 1995)

%z Thesis
%K Prestor01
%A Uros Prestor
%T Evaluating the memory performance of a ccNUMA system
%I Department of Computer Science, University of Utah
%D May 2001

%z Thesis
%K Saavedra92
%A Rafael H. Saavedra-Barrera
%T CPU Performance evaluation and execution time prediction using narrow spectrum benchmarking
%I Department of Computer Science, University of California at Berkeley
%D 1992

%z Article
%K Saavedra95
%A R.H. Saavedra
%A A.J. Smith
%T Measuring cache and TLB performance and their effect on benchmark runtimes
%J IEEE Transactions on Computers
%V 44
%N 10
%D October 1995
%P 1223-1235

%z Article
%k Seltzer99
%A Margo Seltzer
%A David Krinsky
%A Keith Smith
%A Xiolan Zhang
%T The case for application-specific benchmarking
%B Proceedings of the 1999 Workshop on Hot Topics in Operating Systems
%C Rico, AZ
%D 1999

%z Article
%K Shein89
%A Barry Shein
%A Mike Callahan
%A Paul Woodbury
%T NFSSTONE: A network file server performance benchmark
%B Proceedings USENIX Summer Conference
%C Baltimore, MD
%D June 1989
%P 269-275

%z Article
%K Staelin98
%A Carl Staelin
%A Larry McVoy
%T mhz: Anatomy of a microbenchmark
%B Proceedings USENIX Annual Technical Conference
%C New Orleans, LA
%D June 1998
%P 155-166

%z Article
%K FSF89
%A Richard Stallman
%Q Free Software Foundation
%T General Public License
%D 1989
%O Included with \*[lmbench]

%z Book
%K Toshiba94
%A Toshiba
%T DRAM Components and Modules
%I Toshiba America Electronic Components, Inc.
%P A59-A77,C37-C42
%D 1994

%z Article
%K Tullsen96
%A Dean Tullsen
%A Susan Eggers
%A Joel Emer
%A Henry Levy
%A Jack Lo
%A Rebecca Stamm
%T Exploiting choice: Instruction fetch and issue on an implementable simultaneous multithreading processor
%C Proceedings of the 23rd Annual International Symposium on Computer Architecture
%D May 1996
%P 191-202
%O http://www.cs.washington.edu/research/smt/papers/ISCA96.ps

%z Article
%K Tullsen99
%A Dean Tullsen
%A Jack Lo
%A Susan Eggers
%A Henry Levy
%T Supporting fine-grain synchronization on a simultaneous multithreaded processor
%B Proceedings of the 5th International Symposium on High Performance Computer Architecture
%D January 1999
%P 54-58
%O http://www.cs.washington.edu/research/smt/papers/hpca.ps

%z Article
%K Weicker84
%A R.P. Weicker
%T Dhrystone: A synthetic systems programming benchmark
%J CACM
%V 27
%N 10
%P 1013-1030
%D 1984

%z Article
%K Wolman89
%A Barry L. Wolman
%A Thomas M. Olson
%T IOBENCH: a system independent IO benchmark
%J Computer Architecture News
%V 17
%N 5
%D September 1989
%P 55-70
%x IOBENCH is an operating system and processor independent synthetic
%x input/output (IO) benchmark designed to put a configurable IO and
%x processor (CP) load on the system under test.  This paper discusses
%x the UNIX versions.
%k IOBENCH, synthetic I/O benchmark, UNIX workload
%s vinton%cello@hplabs.hp.com (Fri Sep 20 12:55:58 PDT 1991)