For a high-performance parallel machine to be a scal-able system, it must afso have a scalable parallel 1/0 system. This paper presents an experimental evaluation of the Intel Touchstone Delta’s Concurrent File System ( CFS). The main objective of the stu
.An Experimental
Performance Concurrent
Evaluation
of Touchstone
Delta
File System* Juan Hall, 13210 NY
Rajesh
R. Bordawekar NPAC Syracuse
Alok and
N. Choudhary ECE Dept., 121 Link Syracuse,
Miguel
del Rosariot
University,
ABSTRACT For able of the ( CFS). the tions the on the a high-performance system, This Intel The of 1/0 effects that system it must paper Touchstone main and of file parallel afso presents objective compute access performance. of but that of the the lower have machine a scalable Concurrent study for In buffer In CFS than in most scales the the rates nodes. modes, various addition, sizes and cases, to be a scal1/0 evaluation File System is to determine configurawe study file the sizes result number imcaries howsystem that the and peak They
Some ers include are parallel
of the Intel
commercially Paragon[4],
available nCUBE
parallel[8], of in
comput[12]. in form research developed Howthe or
CM-.5
computational community and every and
instruments can major be found
choice one and
scientific another institution. at various ever,
system.
an experimental of the
Delta’s
in
almost Several industrial
academic are also accorded memories has 1/0
prototypes attention and attention
being
maximum
file read/write
academic
institutions.
compared
to the networks very little
to processors, of these paralto the
interconnection lel systems,
been
paid
shows of disks
performance are number and of much
as the performance
scalability Providing without
and
performance raw processing 1/0
of the
system. and large however, memois not A bal-
is increased, If’e observe
sustained performance due
provements pacity. with ever, that data the
system’s beginning, to the 1/0
speed capabilities,
of CFS
scales
balancing to solve many
processors reached software Finally CFS can
a plateu
a quickly enormous the
sufficient anced cessors of high tor
real-world which speeds can
problems. supply to data the
bottleneck performance
overhead, we also show vary
especially for various
memory at the
hierarchy required
to prosuccess
of synchronization. distributions
is critical Although and
greatly employed
performance have
computing. become has not memory developed the 1/0 cheaper kept speeds.
semiconducfaster, the 1/0 sys-
commonly
in scientific
memories
engineering
applications.
tems’
performance and been and
pace Very
with few
advances 1/0 pro-
in processor
parallel the
1
INTRODUCTION the last decade, and this processor trend During been speeds is likely the made High in part
systems cessor
have speeds this
which bandwidth. on the System study
can balance
During
have to
increased for tion stone couprovide In paper, we focus File is to experimental of the the Intel evaluaTouchof the Delta. Concurrent The on the that that goal
tremendously the foreseeable ant
continue
future. have
same
period,
several
import
strides
in high-performance speed processors to few under
effects 1/0
of various system. paWe way.
computing pled with computing “ This work
architectures. massive power was
workloads Based rameters believe
performance study, system can help the
of a parallel we identify in a significant users
parallelism in teraflops sponsored in
are the by
expected next DARPA
on the
experimental affect the study
various
years. contract of the be
to use the important
system input
# DABT63-91-C-0028 and NSF MIP-911081O, The content the information does not necessarily reflect the position or POIICY of the Government and no official endorsement should]nferred. tNpAC and CS Dept Syracuse
in
a better
fashion
as well
as provide systems,
to the The
designers rest
of future paper the
University
of the
is organized parallel 1/0
as follows. model and
In secthe the Intel eval-
tion Permission to copy without fee all or part of this material ie granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its data appear, and notice is given that eop~ing is bv permission of tha Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. ICS-7193 Tokyo, Japan * 1993 ACM 0.8979 j-600-~/93/0007\0367.,.$~ .50 uation
2, we describe Delta
Touchstone
system. and
Section lists some
3 discusses of the
methodologies in the results results the paper paper. for are
experiments performance Array 5. Firs ally, dis. we
presented evaluation tribution conclude
Section various
4 presents experiments. in section 6.
presented in section
367

Delta Routing Improving ... 暂无评价 14页 免费 Evaluation of the commun....File 1/0 performance of the meshattached Concurrent File System is measured....
Delta Routing Improving ... 暂无评价 14页 免费 Evaluation of the commun....File 1/0 performance of the meshattached Concurrent File System is measured....
热门文档