35.2 Introduction

IceGrid is the location and activation service for Ice applications. In prior Ice releases, the IcePack service supplied this functionality. Given the increasing importance of grid computing, IceGrid was introduced in Ice 3.0 to support all of IcePack’s responsibilities and extend it with new features to simplify the development and administration of Ice applications on grid networks.

For the purposes of this chapter, we can loosely define grid computing as the use of a network of relatively inexpensive computers to perform the computational tasks that once required costly "big iron." Developers familiar with distributed computing technologies such as Ice and CORBA may not consider the notion of grid computing to be particularly revolutionary; after all, distributed applications have been running on networks for years, and the definition of grid computing is sufficiently vague that practically any server environment could be considered a "grid."

One possible grid configuration is a homogeneous collection of computers running identical programs. Each computer in the grid is essentially a clone of the others, and all are equally capable of handling a task. As a developer, you need to write the code that runs on the grid computers, and Ice is ideally suited as the infrastructure that enables the components of a grid application to communicate with one another. However, writing the application code is just the first piece of the puzzle. Many other challenges remain:

• How do I install and update this application on all of the computers in the grid?

• How do I keep track of the servers running on the grid?

• How do I distribute the load across all the computers?

• How do I migrate a server from one computer to another one?

• How can I quickly add a new computer to the grid?

Of course, these are issues faced by most distributed applications. As you read this chapter and learn more about IceGrid’s capabilities, you will discover that it offers solutions to these challenges. To get you started, we have summarized IceGrid’s feature set below:

• Location service

As an implementation of an Ice location service (see Section 28.17), IceGrid enables clients to bind indirectly to their servers, making applications more flexible and resilient to changing requirements.

• On-demand server activation

Starting an Ice server process is called server activation. IceGrid can be given responsibility for activating a server on demand, that is, when a client attempts to access an object hosted by the server. Activation usually occurs as a side effect of indirect binding, and is completely transparent to the client.

• Application distribution

IceGrid provides a convenient way to distribute your application to a set of computers, without the need for a shared file system or complicated scripts. Simply configure an IcePatch2 server (see Chapter 42) and let IceGrid download the necessary files and keep them synchronized.

• Replication and load balancing

IceGrid supports replication by grouping the object adapters of several servers into a single virtual object adapter. During indirect binding, a client can be bound to an endpoint of any of these adapters. Furthermore, IceGrid monitors the load on each computer and can use that information to decide which of the endpoints to return to a client.

• Sessions and resource allocation

An IceGrid client establishes a session in order to allocate a resource such as an object or a server. IceGrid prevents other clients from using the resource until the client releases it or the session expires. Sessions enhance security through the use of an authentication mechanism that can be integrated with a Glacier2 router.

• Automatic failover

Ice supports automatic retry and failover in any proxy that contains multiple endpoints. When combined with IceGrid’s support for replication and load balancing, automatic failover means that a failed request results in a client transparently retrying the request on the next endpoint with the lowest load.

• Dynamic queries

In addition to transparent binding, applications can interact directly with IceGrid to locate objects in a variety of ways.

• Status monitoring

IceGrid supports Slice interfaces that allow applications to monitor its activities and receive notifications about significant events, enabling the development of custom tools or the integration of IceGrid status events into an existing management framework.

• Administration

IceGrid includes command-line and graphical administration tools. They are available on all supported platforms and allow you to start, stop, monitor, and reconfigure any server managed by IceGrid.

• Deployment

Using XML files, you can describe the servers to be deployed on each computer. Templates simplify the description of identical servers.

As grid computing enters the mainstream and compute servers become commodities, users expect more value from their applications. IceGrid, in cooperation with the Ice run time, relieves you of these low-level tasks to accelerate the construction and simplify the administration of your distributed applications.