A Scheme for Transfer and Execution of Architecture Independent Procedures

[megha]

[attachment=15242]
Abstract
Conventional Remote Procedure Call (RPC) mechanisms allows processes to pass argu-
ments to a procedure residing in a remote machine, execute it there and obtain results. I have
designed and implemented a scheme which will allow a process to send the arguments for a
procedure to a remote machine for execution. But unlike conventional RPC, the procedure
need not reside on the machine on which it is to be executed. The remote machine fetches
the procedure as and when required from the network (possibly from the client itself). The
procedures (called modules) are stored and transfered in an intermediate format, which is
independent of any specific architecture or operating system. They are translated on the fly
to machine code for execution. MD5 sums over the code of the module is used to identify
modules. These can be used by the remote machine to authenticate a module fetched from
the network. Security issues are handled by using a MD5 sums and a protected address space
provided by the operating system on the remote machine. The scheme also allows a process
to fetch a module from the network to be executed locally in its own address space. Caching
of modules is used to improve performance.
This scheme tremendously increases the flexibility of network applications. If this scheme is
implemented widely, an application will no longer be constrained to use the standard protocols
to communicate to other machines which does not have the application running. Whenever
a standard protocol is not suitable, the application developer can design a new protocol and
create a server program (a module) which can be remotely executed on the server machine.
Module caching ensures that the overheads involved are automatically reduced if the protocol
becomes widely used.