Modern middleware technologies attempt to ease the programmer’s transition to distributed application development by making remote invocations as easy to use as traditional method calls: a method is invoked on an object and, when the method completes, the results are returned or an exception is raised. Of course, in a distributed system the object’s implementation may reside on another host, and consequently there are some semantic differences that the programmer must be aware of, such as the overhead of remote invocations and the possibility of network-related errors. Despite those issues, the programmer’s experience with object-oriented programming is still relevant, and this
synchronous programming model, in which the calling thread is blocked until the operation returns, is familiar and easily understood.
Ice is inherently an asynchronous middleware platform that simulates synchronous behavior for the benefit of applications (and their programmers). When an Ice application makes a synchronous twoway invocation on a proxy for a remote object, the operation’s in parameters are marshaled into a message that is written to a transport, and the calling thread is blocked in order to simulate a synchronous method call. Meanwhile, the Ice run time operates in the background, processing messages until the desired reply is received and the calling thread can be unblocked to unmarshal the results.
There are many cases, however, in which the blocking nature of synchronous programming is too restrictive. For example, the application may have useful work it can do while it awaits the response to a remote invocation; using a synchronous invocation in this case forces the application to either postpone the work until the response is received, or perform this work in a separate thread. When neither of these alternatives are acceptable, the asynchronous facilities provided with Ice are an effective solution for improving performance and scalability, or simplifying complex application tasks.
Asynchronous Method Invocation (
AMI) is the term used to describe the client-side support for the asynchronous programming model. Using AMI, a remote invocation does not block the calling thread while the Ice run time awaits the reply. Instead, the calling thread can continue its activities, and the application is notified by the Ice run time when the reply eventually arrives. Notification occurs via a callback to an application-supplied programming-language object
1. AMI is described in detail in
Section 33.3.
The number of simultaneous synchronous requests a server is capable of supporting is determined by the server’s concurrency model (see
Section 32.9). If all of the threads are busy dispatching long-running operations, then no threads are available to process new requests and therefore clients may experience an unacceptable lack of responsiveness.
Asynchronous Method Dispatch (
AMD), the server-side equivalent of AMI, addresses this scalability issue. Using AMD, a server can receive a request but then suspend its processing in order to release the dispatch thread as soon as possible. When processing resumes and the results are available, the server sends a response explicitly using a callback object provided by the Ice run time.
In practical terms, an AMD operation typically queues the request data (i.e., the callback object and operation arguments) for later processing by an application thread (or thread pool). In this way, the server minimizes the use of dispatch threads and becomes capable of efficiently supporting thousands of simultaneous clients.
An alternate use case for AMD is an operation that requires further processing after completing the client’s request. In order to minimize the client’s delay, the operation returns the results while still in the dispatch thread, and then continues using the dispatch thread for additional work.
See Section 33.4 for more information on AMD.
A programmer indicates a desire to use an asynchronous model (AMI, AMD, or both) by annotating Slice definitions with metadata (
Section 4.17). The programmer can specify this metadata at two levels: for an interface or class, or for an individual operation. If specified for an interface or class, then asynchronous support is generated for all of its operations. Alternatively, if asynchronous support is needed only for certain operations, then the generated code can be minimized by specifying the metadata only for those operations that require it.
Synchronous invocation methods are always generated in a proxy; specifying AMI metadata merely adds asynchronous invocation methods. In contrast, specifying AMD metadata causes the synchronous dispatch methods to be
replaced with their asynchronous counterparts. This semantic difference between AMI and AMD is ultimately practical: it is beneficial to provide a client with synchronous and asynchronous versions of an invocation method, but doing the equivalent in a server would require the programmer to implement both versions of the dispatch method, which has no tangible benefits and several potential pitfalls.
["ami"] interface I {
bool isValid();
float computeRate();
};
interface J {
["amd"] void startProcess();
["ami", "amd"] int endProcess();
};
In this example, all proxy methods of interface I are generated with support for synchronous and asynchronous invocations. In interface
J, the
startProcess operation uses asynchronous dispatch, and the
endProcess operation supports asynchronous invocation and dispatch.
Specifying metadata at the operation level, rather than at the interface or class level, not only minimizes the amount of generated code, but more importantly, it minimizes complexity. Although the asynchronous model is more flexible, it is also more complicated to use. It is therefore in your best interest to limit the use of the asynchronous model to those operations for which it provides a particular advantage, while using the simpler synchronous model for the rest.
The use of an asynchronous model does not affect what is sent “on the wire.” Specifically, the invocation model used by a client is transparent to the server, and the dispatch model used by a server is transparent to the client. Therefore, a server has no way to distinguish a client’s synchronous invocation from an asynchronous invocation, and a client has no way to distinguish a server’s synchronous reply from an asynchronous reply.
