Information Technology Reference
In-Depth Information
Table 7. Results from the bandwidth and throughput measurements
Size of Segment (bytes)
1024
2048
4096
8192
16384
32767
65535
Scenario 1
Throughput
6781.586
3638.966
1794.872
871.818
460.053
230.603
120.902
Std. Dev.
186.270
25.994
37.553
7.248
7.471
3.927
1.467
Bandwidth (MB/s)
6.623
7.107
7.011
6.811
7.188
7.206
7.556
Scenario 2
Throughput
6646.947
3627.420
1773.944
892.972
454.091
235.335
117.091
Std. Dev.
140.815
26.239
29.443
7.959
7.517
2.163
1.622
Bandwidth (MB/s)
6.491
7.085
6.929
6.976
7.095
7.354
7.318
Scenario 3
Throughput
8841.626
3983.438
2166.548
1185.448
575.914
300.792
152.712
Std. Dev.
249.939
15.483
19.652
14.235
8.900
4.880
1.309
Bandwidth (MB/s)
8.634
7.780
8.463
9.261
8.999
9.399
9.544
REFERENCES
is achieved with the caching mechanism and the
asynchronous operation using the in-bound thread.
A communication state transfer mechanism is
employed, which re-establishes connections with
migrated components to improve the data stream
transfer time. Experimental results show that the
caching mechanism is able to buffer channel data
segments and when asynchronous operation is
employed, throughput is substantially increased.
With asynchronous operation, sending and receiv-
ing of messages can be overlapped, resulting in
improved message sending rates than normal
non-asynchronous operation.
Abramson, D., Foster, I., Giddy, J., Lewis, A.,
Sosic, R., Sutherst, R., & White, N. (1997). Nim-
rod Computational Workbench: A Case Study in
Desktop Metacomputing. In Australian Computer
Science Conference (ACSC 97) . Macquarie Uni-
versity, Sydney.
Abramson, D., & Kommineni, J. (2004). A Flex-
ible IO Scheme for Grid Workflows. In Proc. of
the 18th International Parallel and Distributed
Processing Symposium . Krakow, Poland.
Abramson, D., Sosic, R., Giddy, J., & Hall, B.
(1995). Nimrod: A Tool for Performing Parameter-
ised Simulations using Distributed Workstations.
In Proc. of the 4th IEEE Symposium on High
Performance Distributed Computing . Virginia.
IEEE Press.
ACKNOWLEDGMENT
We thank the Victorian Partnership for Advanced
Computing (VPAC) for the use of their facilities
and for their continuing support for this project.
Batchu, R., Dandass, Y. S., Skjellum, A., & Beddhu,
M. (2004). MPI/FT: A Model-Based Approach to
Low-Overhead Fault Tolerant Message-Passing
Middleware. Cluster Computing , 7 (4), 303-315.
doi:10.1023/B:CLUS.0000039491.64560.8a
 
Search WWH ::




Custom Search