Push-Based Prefetching in Remote Memory Sharing System - Cloud, Grid and High Performance Computing: Emerging Applications - page 283

the prediction has a coarse granularity. Choi et al .
present a prediction based on disk blocks(Choi,
et al. , 2000). The algorithm classifies disk ac-
cess into several predefined patterns, and predict
current pattern when accessing disk blocks. The
work of Gniady et al . predicts I/O access using
the program counter (Gniady, et al. , 2004). The
system maintains a hash table from the program
counter to the access pattern and predict the pat-
tern. Different with those previous works, our
system predict the I/O access using a data mining
method, which usually has finer granularity and
better precision.
Our algorithm for collecting the trace and infer-
ring the pattern of users is based on the problem of
sequential pattern mining. Agrawal et al . first de-
fined the problem of sequential patterns (Agrawal,
et al. , 1995, 1996). However, their algorithms are
not applicable to for very long sequences which
are often the case in grid environments. Pei et
al . proposed PrefixSpan algorithm (Pei, et al. ,
2001), which improves upon Apriori and reduces
the overhead. These algorithms are based on the
general problem of sequential pattern mining in
very large databases (Ayres, et al. , 2002); while
the background of our algorithm is quite specific,
some of the restrictions in traditional sequential
pattern mining can be released and the algorithm
is also more effective.

a prefetching algorithm to push more pages to a
user node. By mining the historical information,
a memory node can push the required data to user
nodes efficiently. We demonstrate the efficiency
and effectiveness of the proposed prefetching
scheme through comprehensive trace-driven
simulations.

ACKNOWLEDGMENT

The work was supported by the National
Natural Science Foundation of China under
Grant No.61003076 and the National Basic
Research Program of China (973) under Grant
No.2011CB302600.

REFERENCES

Agrawal, R., & Srikant, R. (1995). Mining se-
quential patterns. Paper presented at the 17th
International Conference on Data Engineering.

Agrawal, R., & Srikant, R. (1996). Mining se-
quential patterns: Generalizations and perfor-
mance improvements. Paper presented at the 5th
International Conference on Extending Database
Technology: Advances in Database Technology.

Ayres, J., Flannick, J., Gehrke, J., & Yiu, T. (2002).
Sequential pattern mining using a bitmap repre-
sentation. Paper presented at the 8th International
Conference on Knowledge Discovery and Data
Mining Edmonton, Alberta, Canada.

CONCLUSION

With the rapid development of the network
technology, several remote memory sharing sys-
tems have been proposed to aggregate memory
resources through definite network environment.
Our previous work, RAM Grid, made use of the
remote memory to boost the performance of
memory-intensive and I/O-intensive applications.
In this paper, in order to reduce the network com-
munication cost of accessing the remote memory,
based on a push strategy and inspired by traditional
sequential patterns mining techniques, we propose

Chang, E., & Garcia-Molina, H. (1999). Medic:
A memory and disk cache for multimedia clients.
Paper presented at the IEEE International Con-
ference on Multimedia Computing and Systems,
Florence, Italy.