![](bgb.png)
sed of different instance types are an interesting way
to execute parallel, load imbalanced applications. In
such a heterogeneous system, tasks with lower com-
putational demands execute on slower but cheaper
machines, while tasks with higher demands execute
on faster but more expensive machines, thus increa-
sing the overall cost efficiency of the application.
In this paper, we introduced the Imbalance Ben-
chmark (ImbBench), whose main purpose is to help
users to profile their environment in terms of the he-
terogeneity of the instances, and to discover oppor-
tunities for heterogeneous clouds. During the eva-
luation of ImbBench we discovered that, depending
on the application imbalance pattern, it is possible to
improve the cost efficiency of the cloud environment
without or with only a small increase of the execution
time. Our results shown that were possible to incre-
ase the cost efficiency up to 63% with less than 7%
of performance reduction. Experiments with the NAS
Parallel Benchmarks showed that these gains can also
be observed with traditional distributed applications.
These results show us that it is possible for users to
take advantage of the heterogeneity offered by cloud
providers.
In the future, we will increase the capabilities of
ImbBench, adding more features, such as a measure-
ment capability of the network performance, memory
operations, and floating point operations, so that the
environmental profile will be more accurate. We will
extend our evaluation to cover a more diverse environ-
ment as well, by using other cloud providers, a private
cloud, and even the instances with variable costs, such
as the AWS Spot instances. Furthermore, we intend to
develop a mechanism feature to help users take advan-
tage of cloud heterogeneity in an automated way, by
analyzing instance options and application behavior
and providing a recommendation of the most suitable
environment for the users.
ACKNOWLEDGEMENTS
This research received funding from the EU H2020
Programme and from MCTI/RNP-Brazil under the
HPC4E project, grant agreement no. 689772. This re-
search received partial funding from CYTED for the
RICAP Project, grant agreement no. 517RT0529. Ad-
ditional funding was provided by FAPERGS in the
context of the GreenCloud Project.
REFERENCES
Bailey, D. H., Barszcz, E., Barton, J. T., Browning, D. S.,
Carter, R. L., Dagum, L., Fatoohi, R. A., Frederick-
son, P. O., Lasinski, T. A., Schreiber, R. S., et al.
(1991). The nas parallel benchmarks. The Internatio-
nal Journal of Supercomputing Applications, 5(3):63–
73.
Blumofe, R. D. and Leiserson, C. E. (1994). Scheduling
multithreaded computations by work stealing. In Sym-
posium on Foundations of Computer Science (FOCS),
pages 1–29.
Carre
˜
no, E. D., Diener, M., Cruz, E. H. M., and Navaux,
P. O. A. (2016). Automatic communication optimi-
zation of parallel applications in public clouds. In
IEEE/ACM 16th International Symposium on Clus-
ter, Cloud and Grid Computing, CCGrid 2016, Cart-
agena, Colombia, May 16-19, 2016, pages 1–10.
Che, S., Boyer, M., Meng, J., Tarjan, D., Sheaffer, J. W.,
Lee, S. H., and Skadron, K. (2009). Rodinia: A ben-
chmark suite for heterogeneous computing. In 2009
IEEE International Symposium on Workload Charac-
terization (IISWC), pages 44–54.
Cheng, D., Rao, J., Guo, Y., Jiang, C., and Zhou, X. (2017).
Improving performance of heterogeneous mapreduce
clusters with adaptive task tuning. IEEE Transactions
on Parallel and Distributed Systems, 28(3):774–786.
Chohan, N., Castillo, C., Spreitzer, M., Steinder, M., Tan-
tawi, A. N., and Krintz, C. (2010). See spot run: Using
spot instances for mapreduce workflows. HotCloud,
10:7–7.
Crago, S. P. and Walters, J. P. (2015). Heterogeneous cloud
computing: The way forward. Computer, 48(1):59–
61.
Dashti, M., Fedorova, A., Funston, J., Gaud, F., Lachaize,
R., Lepers, B., Qu
´
ema, V., and Roth, M. (2013). Traf-
fic Management: A Holistic Approach to Memory
Placement on NUMA Systems. In Architectural Sup-
port for Programming Languages and Operating Sys-
tems (ASPLOS), pages 381–393.
Diener, M., Cruz, E. H. M., Alves, M. A. Z., Alhakeem,
M. S., Navaux, P. O. A., and Heiß, H.-U. (2015a). Lo-
cality and Balance for Communication-Aware Thread
Mapping in Multicore Systems. In Euro-Par, pages
196–208.
Diener, M., Cruz, E. H. M., and Navaux, P. O. A. (2015b).
Locality vs. Balance: Exploring Data Mapping Poli-
cies on NUMA Systems. In International Conference
on Parallel, Distributed, and Network-Based Proces-
sing (PDP), pages 9–16.
Dong, D., Stack, P., Xiong, H., and Morrison, J. P. (2017).
Managing and unifying heterogeneous resources in
cloud environments. In Proceedings of the 7th Inter-
national Conference on Cloud Computing and Servi-
ces Science - Volume 1: CLOSER,, pages 143–150.
INSTICC, SciTePress.
Dongarra, J. J., Luszczek, P., and Petitet, A. (2003). The
linpack benchmark: past, present and future. Con-
currency and Computation: practice and experience,
15(9):803–820.
CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science
258