carla@cs.duke.edu (Carla Ellis) (05/01/91)
Announcing the availability of two recently completed Ph.D. dissertations
at Duke University in the area of multiprocessor operating systems:
Briefly:
David Kotz's dissertation involves file systems for MIMD multiprocessors.
Rick LaRowe's research involves memory management issues
for NUMA shared memory multiprocessors.
Abstracts appended below.
Both of their dissertations are now available as
Duke Technical Reports (ordering information below).
-------------------------------------------
Prefetching and Caching Techniques
in File Systems
for MIMD Multiprocessors
David F. Kotz
dfk@cs.duke.edu
Duke University CS-1991-016
The increasing speed of the most powerful computers, especially
multiprocessors, makes it difficult to provide sufficient I/O
bandwidth to keep them running at full speed for the largest problems.
Trends show that the difference in the speed of disk hardware and the
speed of processors is increasing, with I/O severely limiting the
performance of otherwise fast machines. This widening access-time gap
is known as the ``I/O bottleneck crisis.'' One solution to the
crisis, suggested by many researchers, is to use many disks in
parallel to increase the overall bandwidth.
This dissertation studies some of the file system issues needed to get
high performance from parallel disk systems, since parallel hardware
alone cannot guarantee good performance. The target systems are large
MIMD multiprocessors used for scientific applications, with large
files spread over multiple disks attached in parallel. The focus is
on automatic caching and prefetching techniques. We show that caching
and prefetching can transparently provide the power of parallel disk
hardware to both sequential and parallel applications using a
conventional file system interface. We also propose a new file system
interface (compatible with the conventional interface) that could make
it easier to use parallel disks effectively.
Our methodology is a mixture of implementation and simulation, using a
software testbed that we built to run on a BBN GP1000 multiprocessor.
The testbed simulates the disks and fully implements the caching and
prefetching policies. Using a synthetic workload as input, we use the
testbed in an extensive set of experiments. The results show that
prefetching and caching improved the performance of parallel file
systems, often dramatically.
-------------------------------------------
Page Placement for Non-Uniform Memory
Access Time (NUMA) Shared Memory
Multiprocessors
Richard P. LaRowe Jr.
rpl@cs.duke.edu
Duke University CS-1991-013
Shared memory multiprocessors are attractive because they are
programmed in a manner similar to uniprocessors. The UMA (uniform
memory access time) class of shared memory multiprocessors is the most
attractive, from the programmer's point of view, since the programmer
need not be concerned with the placement of code and data in the
physical memory modules of the machine. Scalable shared memory
multiprocessors, on the other hand, tend to present at least some
degree of nonuniformity of memory access to the programmer, making the
NUMA (non-uniform memory access time) class an important one to consider.
In this thesis, we investigate the role of the operating system's page
placement policy in managing the physical memory of NUMA
multiprocessors. Analytic, simulation, and implementation techniques
are used to study a wide range of issues and problems associated with
managing the placement of virtual pages in the physical memory modules
of the machine. We find that in the NUMA multiprocessor domain, the
placement of pages becomes an important issue affecting performance.
Placement policies strongly biased towards local frame allocation but
willing to use remote frames under certain conditions perform
significantly better than simpler strategies. We also find that page
placement policies that support the movement of virtual pages from
memory module to memory module (page migration) and the creation of
multiple physical copies of a single virtual page (page replication)
dramatically improve the performance of programs written using the UMA
memory model. The dissertation investigates a number of issues
affecting the performance of page placement policies in the face of
different application programs and architectural features.
-------------------------------------------
A copy of either thesis is $5. Please send your request (indicate
which TR you would like) to kathy@cs.duke.edu, or
Kathy Redding
Department of Computer Science
Duke University
Durham, NC 27706 USA
--
=========================== MODERATOR ==============================
Steve Stevenson {steve,fpst}@hubcap.clemson.edu
Department of Computer Science, comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell