NetSolve: A Network Previous: Taking Failures into

Performance

One of the challenges in designing NetSolve was to combine ease of use and excellence of performance. Several factors ensure good performance without increasing the amount of work required of the user. In addition to the availability of diverse scientific packages (as discussed in the preceding section), these factors include load balancing and the use of simultaneous resources.

• Load balancing. Given all the computational resources available, NetSolve provides the user with a ``best effort'' to find the most suitable resource for a given problem.

•   Simultaneous resources. Using the nonblocking interfaces to NetSolve, the user can write a NetSolve application that has some parallelism. In Figure 4, we see the result of experiments conducted on a network of SPARC workstations. The NetSolve program kept sending requests so that ten eigenvalues problems were solved simultaneously over the network. We also added computational servers to the NetSolve configuration while running this program. Figure 4 shows the execution time for each problem for each experiment. As expected, the problems are solved simultaneously on different servers, and the average execution time for one problem decreases when the number of computational servers increases.

 
Figure 4: Simultaneous request to an evolving NetSolve system