A basic installation of Office SharePoint Server 2007 for Search installs all components on a single server. This means that the index and query server roles run on one computer. This installation option does not support deploying separate query servers because you cannot add servers when deploying using the basic installation option.

We recommend that a basic installation be used only for evaluation purposes. For performance, scale, availability and security reasons, a farm installation is the recommended choice for a production deployment.

When deploying Office SharePoint Server 2007 for Search in a server farm, the following two options are available:

• Option 1   Use the same computer as the query server and index server.

• Option 2   Use one computer as the index server and one or more different computers as query servers. The index server does not serve queries when using this option.

For both options you can have only one index server.

To determine which topology option is best for your organization, consider how many search queries your organization needs to support, and consider the availability requirements for search. You can then balance that information with the cost of deploying additional servers in your server farm.

For organizations that anticipate a large number of queries and have the budget to add servers to their farm, option 2 is typically the best choice. This topology option increases performance because the computer resources that are used to index content are not consumed by servicing search queries. As the number of search queries increases for your server farm, you can add query servers to your server farm to balance the load. If you start with option 1, you must turn off the query role on the original index/query server when adding query servers to the farm.

Note   When one server in the server farm is configured as both an index and query server, the index is not propagated. Instead, queries are served from the content index on the index server. When multiple query servers are deployed, search queries are automatically load balanced between query servers.

Availability is also a factor when deciding how many query servers you should deploy. When multiple query servers are deployed, the OSearch service continually propagates the content index to all query servers as new content is crawled and indexed. Automatically propagating updates to the content index on the index server to the query servers is called continual propagation.

Having separate query servers enables each query server to serve search queries from its own copy of the content index. In server farms where multiple query servers are deployed, if a query server fails, all queries are diverted to other query servers.

Even deployments that have only one query server that is separate from the index server can benefit from the query server having its own copy of the content index. In this server topology, if the index server has a catastrophic failure, the query server continues to serve search queries while the index server is unavailable. Conversely, if the query server has a catastrophic failure in this topology, a farm administrator can temporarily configure the index server to serve queries from the index server's content index until the query server is repaired. Note that after the query server is repaired, the entire content index must be propagated from index server to the query server. The time required to do this varies depending upon the size of your content index. A content index can contain several gigabytes of data, depending on how much content is crawled.

Tip:
Organizations that require a high percentage of uptime for search queries should consider deploying at least two query servers.

Server farms that deploy the index server and query server role on the same computer do not benefit from the redundancy of having multiple copies of the content index. If this single computer has a catastrophic failure, the server farm cannot crawl and index content or serve queries until the problem with the server is resolved.