Search model 3 of 4
Physical and logical components
Search in Microsoft® SharePoint® Server “14” is rearchitected with new components to create greater redundancy within a single farm and to allow scalability in multiple directions. Each of the components that make up the query architecture and the crawling architecture canbe scaled out separately based on the needs of an organization.
Scale-out decision points
Volume of items
How the number of items in your corpus affects farm sizing and the decision to scale out.
Number of Items Action 0-1 million All Search roles can coexist on one or two servers, as shown in the Limited Deployment architecture example. Move the index server role to another server, whilethe query server role remains on the Web servers, as shown in the Small Farm architecture example. Add an index server. Each index server has one crawler. Create another index partition and distribute instances of each index partition to the other query server, as shown in the Medium Shared Farm architecture example. Add query servers and index partitions, with distributed index partition instances.Add another crawl database, associated with a new crawler on each index server. Also, add another property database to the second database server, and mirror both property databases, as shown in the Medium Dedicated Farm architecture example. Isolate each topology layer into "server groups" in which each role is deployed to its own servers. Each server group can then be scaled out to meetspecific requirements for the components in that role.
· · · · Hosts Search Web Parts and Web Part pages for answering search queries. In dedicated search service farms, this role is not necessary because Web servers at remote farms contact query servers directly. This role is necessary for farms that include other SharePoint Server “14” capabilities. In small farms, thisrole can be shared on a server with the query role.
Key performance characteristics and how they relate to scaling of specific search components.
To improve this metric… Take these actions Full crawl time and result Add index servers, crawlers, and crawl freshness databases. Each crawl database can contain content from independent sources. Each crawldatabase can have several crawlers associated with it, and those crawlers can be distributed among many index servers. If you have several content sources, multiple crawlers and associated crawl databases allow you to crawl the content concurrently. Time required for results If query latency is caused by high peak query to be returned load, add query servers and index partitions. Each index partitioncan contain up to ~10 million items, and can also be “cloned” by deploying multiple instances of a given index partition across several query servers. Query throughput increases when you add index partition instances. If query latency is caused by database load, isolate the property database from crawl databases by moving it to a separate database server. Deploy redundant query servers, redundantindex partitions and partition instances, and use clustered or mirrored database servers to host crawl and property databases. Use multiple crawlers on redundant index servers, and add crawl databases. Crawlers associated with a given crawl database can be distributed across index servers for availability and load distribution.
Redundancy and availability
Which search components are redundantand how the redundancy/failover mechanism works.
Component Can be How redundancy or failover works deployed to multiple servers? Yes Each index partition contains a discrete portion of the corpus, and can contain up to 10 million items. Each index partition can be “mirrored” using index partition instances (see below). We recommend that you deploy one query server for each index partition you...