Information Technology Reference
In-Depth Information
5.
Conclusion
In this chapter, we discuss the sabotage-tolerance problems in VC systems which has been
a mandatory issue for high performance and reliable VC environment.
In section 2, we introduced major VC projects/platforms and proposed some practi-
cal models of VC environment. In section 3, we introduced existing sabotage-tolerance
mechanisms and provided their performance evaluation in the practical models. From this
evaluation, we concluded that credibility-based voting is the most promising approach for
sabotage-tolerance of VCs. This section also showed that the performance largely depends
on the job scheduling method in credibility-based voting. Then, in section 4, we proposed
expected-credibility-based job scheduling method for credibility-based voting. The exten-
sive simulations showed that the proposed method reduced the computation time compared
to the original round-robin method, irrespective of the value of unknown parameters.
In the process of applying the proposed job scheduling method to real VCs, some ques-
tions may arise with respect to assumptions or implementation details. Some promising
research that can be further explored , for example, include:
•
Workers' processor speed:
All workers assumed to have the same property, e,g. the same speed, in VC models.
Since each worker in real VC has different property, there may be a big difference
between the simulation results and real environments. Additional simulation will
show the actual performance in real VCs. For such simulation, we can still handle
workers who have different property from others by simply treating them as multiple
workers. For example, a worker with 2GHz processor can be assumed as two workers
with 1GHz processors.
•
Another form of collusion:
The predefined colluding is assumed in the simulation. The predefined colluding
method is the most simple model of collusion. Since saboteurs in this model always
return incorrect results with constant probability
s
, it is easy to detect the collusion
by spot-checking. Saboteurs in real VCs may behave more artfully, e,g, they collude
only when they receive the same job each other since the collusion will succeed with a
higher probability. Such saboteurs can be described using the direct colluding model.
References
[1] Luis F. G. Sarmenta, “Volunteer Computing”, Ph.D. thesis, MASSACHUSETTS IN-
STITUTE OF TECHNOLOGY, 2001.
[2] David M. Toth, “Improving the Productivity of Volunteer Computing”, Ph.D. thesis,
WORCESTER POLYTECHNIC INSTITUTE, 2008.
[3] Thomas E. Anderson, David E. Culler, David A. Patterson, and the NOW Team, “A
case for networks of workstations: NOW”, IEEE Micro, Feb. 1995.