There are a number of issues that must to be dealt with before ParaWeb or other similar approaches will achieve wide-spread acceptance and use. However, many of these issues are not specific to ParaWeb but relate instead to the general issue of executable content over the Web. This includes, for instance, issues of security. On the surface, it would seem that there is an inherent danger in allowing clients to upload and execute programs on a server. However, the risk is no different than when applications are automatically downloaded and executed by clients browsing the Web. Thus, the inherent security features of Java should provide the same level of protection from malicious users. In addition, many computing sites may also likely restrict upload access to trusted users (for instance, users within the organization, or users that have previously registered with the site), thereby limiting the security risk.
Another issue, related to the performance of interpreted Java byte-code, is also beyond the scope of our own work. We expect that the introduction of just-in-time compilers will greatly reduce the performance gap for the types of compute intensive programs which are the target of ParaWeb. Short running programs may not necessarily benefit significantly from just-in-time compilation, but these programs are not likely to be either parallelized or executed remotely anyway.
Authentication and authorization are two other aspects of distributed and parallel computing that need to be further considered if resources spread throughout the Web are to be utilized in a safe and seamless fashion. We are not currently trying to develop solutions to these issues. The recent addition of pickling and remote method invocation (both are briefly described in Section 6) to version 1.1 of the JDK leads us to believe that the demand for authentication and authorization may warrant their addition to future JDK's.