3. Initial Deployment Guidelines
It can be a difficult process to decide how
large of a deployment and how many resources to dedicate to an initial
search deployment. Generally speaking, if SharePoint 2010 is being
installed as the first document management and collaboration tool, the
search components can start out in their simplest form and scale out as
documents are added and the adoption of SharePoint 2010 grows. However,
this scenario is becoming rare, and most organizations will find
themselves either upgrading to SharePoint 2010 from a previous version
or migrating content from another system or from an ad hoc system, such
as a file share. In these cases, initial deployment architectures and
best practices can be useful.
Luckily, Microsoft gives some guidelines for
what kinds of architectures can support different scenarios. These
guidelines are based on the number of items that are indexed and
searchable regardless of whether they come from inside or outside
SharePoint 2010. The entirety of these documents is known as the corpus.
- 1 million documents or less: All roles on a single server or one application server and one database server
- 1 to 10 million documents: Web server and query server roles combined on one to two servers, one dedicated crawl server, and a dedicated database server
- 10–20 million documents: Same architecture as 1 to 10
million documents model but with an additional crawl server and
redundancy on the query servers—that is, use two query servers, each
with two index partitions broken into half of the index partition on
each and a mirror of the other server's half on each.
- 20–40 million documents: For more than 20 million documents,
a dedicated search farm is recommended. On a dedicated search farm,
servers with the web server role are not necessary as long as the farm
where searches are initiated has servers with this role and those
servers are configured to direct queries to the dedicated farm. The
farm should have a minimum of four query servers, each with a quarter
of the index in an index partition. Each query server should also host
a mirror of another server's partition for redundancy. In addition, two
crawl servers, each with two crawlers, and two crawl databases, each
with two crawlers associated with it, are recommended. In addition, it
is recommended that two additional database servers be utilized to hold
property databases and the Search Admin database.
- 40–100 million documents: Any organization with more than 40
million documents will require a completely dedicated search farm with
a high level of redundancy. The web server role should probably be
handled by the content farm except if a dedicated web server is
allocated for indexing (recommended). The farm should have six to ten
query servers, each holding an equal portion of the index in an index
partition, as well as a mirror of one of the other index partitions
from another server. There should also be four crawl servers and four
database clusters where the four crawl servers write to two crawl
databases.
Use the following flowco easily see how many servers will be required for a deployment.
Figure 2. Deployment sizing decision flowchart
