| 1 |
% BibTeX bibliography file |
| 2 |
|
| 3 |
@InProceedings{kotz:addrtrace, |
| 4 |
author = {David Kotz and Preston Crow}, |
| 5 |
title = {The Expected Lifetime of ``Single-Address-Space'' Operating |
| 6 |
Systems}, |
| 7 |
booktitle = {Proceedings of the 1994 ACM Sigmetrics Conference on Measurement |
| 8 |
and Modeling of Computer Systems}, |
| 9 |
year = {1994}, |
| 10 |
month = {May}, |
| 11 |
pages = {161--170}, |
| 12 |
URL = {file://cs.dartmouth.edu/pub/CS-papers/Kotz/kotz:addrtrace.ps.Z}, |
| 13 |
keyword = {operating system, wide address space, virtual memory, memory |
| 14 |
management, dfk}, |
| 15 |
abstract = {Trends toward shared-memory programming paradigms, large (64-bit) |
| 16 |
address spaces, and memory-mapped files have led some to propose the use of a |
| 17 |
single virtual-address space, shared by all processes and processors. Typical |
| 18 |
proposals require the single address space to contain all process-private |
| 19 |
data, shared data, and stored files. To simplify management of an address |
| 20 |
space where stale pointers make it difficult to re-use addresses, some have |
| 21 |
claimed that a 64-bit address space is sufficiently large that there is no |
| 22 |
need to ever re-use addresses. Unfortunately, there has been no data to |
| 23 |
either support or refute these claims, or to aid in the design of appropriate |
| 24 |
address-space management policies. In this paper, we present the results of |
| 25 |
extensive kernel-level tracing of the workstations in our department, and |
| 26 |
discuss the implications for single-address-space operating systems. We found |
| 27 |
that single-address-space systems will not outgrow the available address |
| 28 |
space, but only if reasonable space-allocation policies are used, and only if |
| 29 |
the system can adapt as larger address spaces become available.} |
| 30 |
} |
| 31 |
|
| 32 |
@InProceedings{kotz:pools, |
| 33 |
author = {David Kotz and Carla Ellis}, |
| 34 |
title = {Evaluation of Concurrent Pools}, |
| 35 |
booktitle = {Proceedings of the Ninth International Conference on Distributed |
| 36 |
Computer Systems}, |
| 37 |
year = {1989}, |
| 38 |
pages = {378--385}, |
| 39 |
keyword = {dfk, concurrent pool, concurrent data structure}, |
| 40 |
comment = {also Duke TR CS-1987-30, kotz:poolsTR} |
| 41 |
} |
| 42 |
|
| 43 |
@PhdThesis{kotz:thesis, |
| 44 |
author = {David Kotz}, |
| 45 |
title = {Prefetching and Caching Techniques in File Systems for {MIMD} |
| 46 |
Multiprocessors}, |
| 47 |
year = {1991}, |
| 48 |
month = {April}, |
| 49 |
school = {Duke University}, |
| 50 |
note = {Available as technical report CS-1991-016.}, |
| 51 |
URL = {file://cs.duke.edu/dist/theses/kotz/kotz.ps.Z}, |
| 52 |
keyword = {dfk, parallel file system, prefetching, MIMD, disk caching, |
| 53 |
parallel I/O, pario bib}, |
| 54 |
abstract = {The increasing speed of the most powerful computers, especially |
| 55 |
multiprocessors, makes it difficult to provide sufficient I/O bandwidth to |
| 56 |
keep them running at full speed for the largest problems. Trends show that |
| 57 |
the difference in the speed of disk hardware and the speed of processors is |
| 58 |
increasing, with I/O severely limiting the performance of otherwise fast |
| 59 |
machines. This widening access-time gap is known as the ``I/O bottleneck |
| 60 |
crisis.'' One solution to the crisis, suggested by many researchers, is to |
| 61 |
use many disks in parallel to increase the overall bandwidth. This |
| 62 |
dissertation studies some of the file system issues needed to get high |
| 63 |
performance from parallel disk systems, since parallel hardware alone cannot |
| 64 |
guarantee good performance. The target systems are large MIMD multiprocessors |
| 65 |
used for scientific applications, with large files spread over multiple disks |
| 66 |
attached in parallel. The focus is on automatic caching and prefetching |
| 67 |
techniques. We show that caching and prefetching can transparently provide |
| 68 |
the power of parallel disk hardware to both sequential and parallel |
| 69 |
applications using a conventional file system interface. We also propose a |
| 70 |
new file system interface (compatible with the conventional interface) that |
| 71 |
could make it easier to use parallel disks effectively. Our methodology is a |
| 72 |
mixture of implementation and simulation, using a software testbed that we |
| 73 |
built to run on a BBN GP1000 multiprocessor. The testbed simulates the disks |
| 74 |
and fully implements the caching and prefetching policies. Using a synthetic |
| 75 |
workload as input, we use the testbed in an extensive set of experiments. The |
| 76 |
results show that prefetching and caching improved the performance of |
| 77 |
parallel file systems, often dramatically.}, |
| 78 |
comment = {Published as kotz:prefetch, kotz:jwriteback, kotz:jpractical, |
| 79 |
kotz:fsint2.} |
| 80 |
} |
| 81 |
|