Slide 40 of 47
Notes:
We evaluated the Java Market with respect to nine measures of performance. The Java Market prototype performed very well with respect to seven of them:
* Awareness of changing resource availability.
* Ability to handle resource heterogeneity.
* Security.
* Scalability.
* Transparency (how much extra work the user must do to use the system.)
* Ability to impose a Cost-Benefit based scheduling policy.
* Ability to complete large parallel jobs quickly.
However, due to the limits of web technology, it performed poorly in two:
* Quality of service guarantees.
* Efficient use of resources.
The Java sandbox, even as it provides security, makes it impossible for the Java Market to use the full power of the producer machines. This slows the Market’s performance significantly.
The basic Java assumptions regarding flow of control make it hard to relocate jobs once started. It is similarly difficult to checkpoint jobs to prevent loss of information. For this reason, and the above, we concluded that web technology might not be the right medium for a Java-based metacomputer.
