Yu Xu (yxu@cs.ucsd.edu)
Tue, 16 May 2000 02:03:15 -0700

Evaluation of “ Implementing Remote Procedure Calls“

The authors have implemented a package providing a RPC facility. So, in this paper, they describe the options they have and the decisions they made.
Their main goals include: easy distributed computation, highly efficient RPC communication, secure communication with RPC.
The program structure is base on stubs. There are five kinds of pieces of program in their RPC system: the user, the user stub, the RPC communication package, the server -stub, and the server.
There are several novel aspects in their design.
1. They make the semantics of RPC as close as possible to those of a local procedure call. For example, there are no timeouts mechanism in their RPC.
2.security in the RPC.
3.integration of remote calls into existing programming systems.
4.powerful binding mechanism with the support from Grapevine's distributed database.
5.effiecient transport protocol aiming to reduce the elapsed real-time between initiating a call and getting results and minimize the load imposed on a server by substantial numbers of users.

Evaluation of “ Performing Remote Operations Efficiently on a Local Computer Network”

The paper proposes a Remote reference/Remote Operation model to reduce the traditionally high processing overhead of network communication. Then, they presents implementation considerations for a communication system based on the model and show the implementations for only-once-type-1, primary, processor-synchronous references using request/response protocol.
There are several features of the model.
1. there are two classes of processes: regular processes and recoverable processes. Particularly, Communication processes are recoverable and specially implemented-off-loaded or at interrupt level for quick activation.
2. References have rich attributes. For example, reliability attributes alone include: maybe, at-least-once, only-once-type-1, only-once-type-2.
3. References are divided into three synchrony classes : processor-synchronous, process-synchronous and asynchronous.
4.Operations are divided into two classes: primary and secondary.
The case study results show that the performance of the microcoded implementation is about 30 times faster than the software implementation.
Since in modern RICS machines, microcode are rarely used, the part on microcode in the paper is not important.