home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Language/OS - Multiplatform Resource Library
/
LANGUAGE OS.iso
/
oper_sys
/
quartz
/
quartz10.lha
/
ANNOUNCE
next >
Wrap
Text File
|
1990-05-18
|
2KB
|
47 lines
Announcing the public availability of Quartz, an aid to
tuning parallel programs. Quartz currently runs on the
Sequent Symmetry. It is integrated with Presto, a C++ parallel
programming system, and FastThreads, a lightweight thread package
for C programs.
Quartz are available by anonymous ftp from cs.washington.edu.
A paper about Quartz is to appear at Sigmetrics '90; the abstract
is re-printed below.
Questions and comments are certainly welcome; send them to
tom@cs.washington.edu. Since this is part of an ongoing research
project, also send me mail if you grab a copy and would like to
get updates.
Tom Anderson
---------------------
Quartz: A Tool for Tuning Parallel Program Performance
Thomas Anderson and Edward Lazowska
Quartz is a new tool for tuning parallel program
performance on shared memory multiprocessors.
The philosophy underlying our work is that
an effective tool for tuning parallel program performance
must be based on a clear view of the causes of poor performance,
and on a specific methodology for improving that performance.
By being selective about what it measures and presents, the tool can
focus the programmer's attention on the information needed to tune
performance, eliding details about second-order effects.
Measurement efficiency also is improved by designing the tool to
record just the important behavior. This philosophy is even more
important in the parallel domain than in the sequential domain,
because of the dramatically greater number of possible metrics and
the dramatically increased complexity of program structures.
The principal metric of Quartz is normalized processor time:
the total processor time spent in each section of code divided
by the number of other processors that are concurrently busy when
that section of code is being executed. Tied to the logical structure
of the program, this metric provides a "smoking gun" pointing
towards those areas of the program most responsible for
poor performance. This information can be acquired efficiently
by checkpointing to memory the number of busy processors and the
state of each processor, and then statistically sampling these
using a dedicated processor.