Introduction

So far, the examples we have worked with have been run on the same workstation or host. However, as we gain expertise with interprocess communication techniques, it becomes evident that there will be many occasions when we will want to communicate with processes that may reside on different workstations. These workstations might be on our own local area network or part of a larger wide area network. In a UNIX-based, networked computing setting, there are several ways that communications of this nature can be implemented. This chapter examines the techniques involved with remote procedure calls (RPC). [1] As a programming interface, RPCs are designed to resemble standard, local procedure (function) calls. The client process (the process making the request) invokes a local procedure commonly known as a client stub that contains the network communication details and the actual RPC. The server process (the process performing the request) has a similar server stub , which contains its network communication details. Neither the client nor the server needs to be aware of the underlying network transport protocols. The programming stubs are usually created using a protocol compiler, such as Sun Microsystems rpcgen . This chapter is based on Sun's RPC implementation as ported to Linux. The protocol compiler is passed a protocol definition file written in a C-like language. For rpcgen , the language used is called RPC language. The protocol definition file contains a definition of the remote procedure, its parameters with data types, and its return data type.

[1] The word remote in RPC is somewhat misleading. RPCs can also be used by processes residing on the same host (indeed, this approach is often used when debugging routines that contain RPCs).

When the client invokes an RPC (generates a request), the client waits for the server to reply. Since the client must wait for a response, several coordination issues are of concern:

• How long should a client wait for a reply from the server (the server could be down or very busy)? In general, RPCs address this problem by using a default timeout to limit the client's wait time.
• If the client makes multiple, identical requests , how should the server handle it? The resolution of this problem proves to be program-specific. Depending upon the type of processing (such as a read request), the requested activity may indeed be done several times. In other settings, such as transaction processing, the request must be done only once. In these settings, the software must implement its own management routines. By definition, RPCs are independent of transport protocols; however, if an RPC runs on top of a reliable transport (such as TCP), the client can infer from receiving a reply from the server process that its request will be executed.
• How can call-by-reference (the passing of address pointers) be implemented when the processes reside in separate process spaces? Further, it is entirely possible that the client and server processes, while not being on the same system, may even be executing on different platforms (e.g., Sun, VAX, IBM, etc.). To resolve these issues and to ensure that client and server processes can communicate using RPC, the data that is passed between the processes is converted to an architecture-independent representation. The data format used by Sun is known as XDR (e X ternal D ata R epresentation). The client and server program stubs are responsible for translating the transmitted data to and from the XDR format. This process is known as serialization and deserialization . The high-level relationships of client and server processes using an RPC are shown in Figure 9.1.

  Figure 9.1. An RPC clientserver communications overview.

We will find that, while hidden from the casual user , RPC uses socket-based communication. The details of socket-based communication are covered in Chapter 10, "Sockets."

At a system level, the rpcinfo command can be used to direct the system to display all of the currently registered RPC services. When the -p flag is passed to rpcinfo , the services for the current host are displayed. The rpcinfo command is often located in /usr/sbin . If this is the case on your system, /usr/sbin should be part of the search path (so you will not have to fully qualify the path for each invocation). Some versions of rpcinfo require the host name to be specified. If you do not know the host name , the hostname command will display the name of the host upon which you are working.