Supporting the development of network programs

predictable than that of more traditional distributed memory systems, such as hypercubes [22], since both the For computation-intensive computations, or applications nodes and the network are shared with other users. The with heterogeneous components, partitioning the unpredictable nature of the network environment makes application over a group of computer systems connected the already difficult task of programming distributed-by a network is an attractive solution. Unfortunately, memory computers even more complicated. programmers who want to do ''network computing'' face several challenges: the network and attached systems are The dynamic nature of network computers often shared resources with an unpredictable behavior and requires programmers to use a more complex network communication primitives are often hard to use. programming model when parallelizing their application. The programming environment developed for the Nectar For example, for a lot of applications one can achieve system addresses both problems. It provides simple and good load balancing in a static, homogeneous efficient communication primitives, and an efficient environment, by distributing the input evenly across the monitoring kernel that allows both programmers and nodes (domain decomposition). In a network programming tools to monitor the behavior of the environment however, nodes cannot only have different program in the dynamic network environment. Our speeds, but the load on the nodes can change during experience shows that monitoring the progress of execution. If the data is distributed statically, it might be applications interactively is both desirable and practical. necessary to redistribute the data if the load on the nodes changes, or alternatively, the distribution of the work can be done completely at runtime. Depending on the