Internet Sites Search Architectures
for SharePoint Server 2016
Overview Search topologies for Internet Sites
This model illustrates a medium-size Internet Sites farm architecture. Architecture Topologies for Internet Sites differ from other search topologies in the following ways:
requirements can vary depending on the composition of the data that is crawled (size of Web servers are incorporated in the topology for Internet Sites to serve user requests. This is different in search topologies for Enterprise
items and formats). This example illustrates the type of search components needed and Search where Web servers are not part of the dedicated search farm.
how many of each. Use this example as a starting point for planning your own search The query processing component and index components reside on Web servers to make maximum use of the available hardware
environments. For more information about search processes and how search components resources and to simplify scaling out the search topology.
interact, see Search Architectures for SharePoint® Server 2016 (https://go.microsoft.com/
fwlink/p/?linkid=832554).
Search components Hardware requirements
Index
Index component These requirements apply to the medium Internet Sites search topology. The minimum listed RAM requirements for a server that hosts a
The index component is the logical representation of an index replica. search component is the total required amount of RAM for that server. For example, if you are hosting a content processing component, a
search administration component and a crawl component on one server, the total amount of minimum required RAM for that server is 24
Index partitions GB.
You can divide the index into discrete portions, each holding a separate part of the
index.
An index partition is stored in a set of files on a disk.
The search index is the aggregation of all index partitions. Minimum hardware requirements for application servers
Index replicas
SEARCH COMPONENT ON THE RAM HARD DISK PROCESSOR
Each index partition holds one or more index replicas that contain the same
PHYSICAL SERVER
information. 48 GB for each server in the farm that ALL COMPONENTS:
You have to provision one index component for each index replica. hosts an index component, a query 500 GB additional disk space, preferably a separate disk
Index component
To achieve fault tolerance and redundancy, create additional index replicas for each processing component, and the Web volume/partition. 64-bit, 4 cores minimum, 8 cores
front-end. recommended.
index partition and distribute the index replicas over multiple application servers. 24 GB for each server in the farm that
300 GB additional disk space, preferably a separate disk
hosts an analytics processing
volume/partition. This disk space is necessary for local
Analytics processing component component, a crawl component,
processing of analytics data before it is written to the
Query processing and/or a search administration
analytics reporting database.
component.
Query processing component Crawl component
Analyzes and processes search queries and results. 80 GB for your system drive.
You must have sufficient space for the base installation
See the requirements listed for the
Content processing component
analytics processing component. and sufficient space for diagnostics such as logging and
Admin debugging, for creating memory dumps, and for other
operations. For production use, you also need additional
Search administration component free disk space for regular operations. Maintain twice as
See the requirements listed for the
Runs system processes that are essential to search. There can be more than one search Query processing component
index component.
much free space as you have RAM for production
environments.
administration component per Search service application, but only one component is Search administration component
See the requirements listed for the
active at any given time.
analytics processing component.
Crawl
Crawl component Minimum hardware requirements for database servers
Crawls content based on what is specified in the crawl databases.
HARDWARE COMPONENT REQUIREMENTS
Processor 64-bit, 4 cores for small deployments
Content processing 8GB for small deployments
RAM
Content processing component 16GB for medium deployments
Hard Disk 80GB for system drive
Carries out various processes on the crawled items, such as document parsing and
property mapping.
Analytics
Analytics processing component
Carries out search analytics and usage analytics.
Search databases Medium Internet Sites farm (~85 Page views per
Search admin DB second)
Search administration database
Stores search configuration data. Only one search administration database per
Search service application. This farm is intended to provide a fault-tolerant SharePoint Server 2016 search farm topology that is optimized for a corpus that contains
3,400,000 items. The example farm processes 100-200 documents per second, depending on the language, and it accommodates 85 page
Crawl DB views per second and 100 queries per second.
Crawl database This topology is optimized for physical hardware, but you can deploy it on virtual machines as well.
Stores the crawl history and manages crawl operations. Each crawl database can
have one or more crawl components associated with it.
Host A Host B Host C
Web Servers
Link DB Web Server Web Server Web Server
Web front-end Web front-end Web front-end
Link database Query processing Query processing Query processing
Stores the information extracted by the content processing component and also Replica Index partition 0 Replica Replica
stores click-through information.
Analytics DB
Host
Analytics reporting database To scale out: add 1 Web server with a
Web front-end, query processing
Stores the results of usage analytics. Web Server
component and index component
Web front-end representing a replica to allow for an
Query processing
additional 28 page views per second.
Replica
Performance considerations
Host D Host E Host F
Application Servers
Application servers hosting search components Application Server Application Server Application Server
Crawl Crawl Crawl
Admin Admin Content processing
Analytics Content processing
THINGS TO CONSIDER REASONING Content processing
The query and its results are cached with Windows Server AppFabric,
in key-value pairs: the query being the key and the results being the
value. For each query there is an approximate 50% cache ratio. This
Host
means that if you have a usage pattern of 200 queries per second,
Cache To scale out: add 1 Application server
about 100 queries will be sent to the search index and the other 100 Application Server with a crawl component and a content
processing component to process an
queries will be cached. Results from the cache have lower query Crawl
additional 40 documents per second.
latency than those from the search index. For example, results for Content processing
front-page queries that are run repeatedly are likely to be cached.
We recommend that you enable continuous crawl with an inteval of
Continuous crawl 1 minute instead of the default interval of 15 minutes. You can
enable continuous crawl on SharePoint content sources only.
With anonymous access, users don't have to use credentials to log Host G Host H
Database Servers
into a SharePoint Internet site. In addition, anonymous queries are
All SharePoint Databases All SharePoint Databases
Anonymous access cached, so they are cheaper because of lower query latency. You
must enable anonymous access in two locations: on the web front- Search admin DB Redundant copies of all
databases using SQL
end and on the site itself. Link DB
clustering, mirroring, or
Query latency is influenced by caching, anonymous access, and by Crawl DB SQL Server 2012
AlwaysOn
other factors such as the number and complexity of query rules that Crawl DB
Query latency are applied and triggered. Also, consider the disks on which the Analytics DB
search index is stored; a disk that has multiple spindles can improve All other SharePoint
Databases
the access speed of the disk and reduce query latency.
Paired hosts for fault tolerance
© 2016 Microsoft Corporation. All rights reserved. To send feedback about this documentation, please write to us at [email protected].
