Detailed Recommendations

Microsoft conducted extensive testing of SharePoint Portal Server to ensure that you can perform detailed planning prior to deployment and to ensure that deployment plans deliver a portal that truly makes users more effective. It is not possible to characterize all server performance without providing an extraordinary amount of data. To help customers focus on the most relevant data, the testing efforts focused on scenarios that experience has shown to be common.

The detailed capacity planning recommendations fall into three broad categories:

• Corpus characteristics. Corpus limits describe maximum recommended values for several important characteristics such as the number of documents per workspace, etc.
• Maximum throughput. Maximum throughput results describe the expected optimal throughput for a given server. This information can help you plan an optimal deployment.
• Resource consumption. Resource consumption recommendations allow you to make detailed predictions of the future server requirements for CPU resources, RAM, and hard disk storage.

Identifying Corpus Characteristics

Each of the recommended limits in this section was obtained by starting with the test workspace shown and modifying it with the characteristic in question until end-user latencies were no longer acceptable.

The base workspace for the tests had the following characteristics.

Workspace Characteristics
Number of documents	50,000
Number of document versions	100,000
Average number of versions per document	2
Document profiles	50
Categories	75
Number of documents per folder or category	Up to 20
Number of categories per document	Up to 3
Average number of documents per folder or category	10
Average number of role principals	20
Average file size	100 kilobytes (KB)
Number of subscriptions	1,000
Number of subscription results	2,000

Summary of Recommendations

The following table displays a summary of the recommendations presented in this section. These recommendations do not represent the limit for support from Microsoft Product Support Services. The recommendations can help you plan your deployment. Your specific requirements may vary, and you should adequately test performance and usability. For details and caveats regarding these limits, read the sections that follow this table.

Recommendations
Maximum number of documents stored per server	500,000
Maximum number of document versions per server	1,000,000
Maximum number of documents included in the index per server	3,500,000
Maximum number of documents per folder	200–3,0001
Maximum document size	50 MB
Maximum number of versions per document	1,000
Maximum number of subscriptions per workspace	100,000
Maximum number of document profiles	500
Maximum number of content sources	100
Maximum number of role principals per folder	150–600
Maximum number of categories	500

The following list provides more detail about the recommended maximums.

• Maximum number of documents stored per server. The maximum recommended number of documents stored per server is 500,000, with 1 million document versions (an average of two versions per document). This number is the same regardless of the number of workspaces on the server. Therefore, if a workspace requires storage for 1 million document versions, then it should be the only workspace on the server.
• Maximum number of documents included in the index per server. The maximum recommended number of documents included in the index per server is 3.5 million. The primary resource constraint for this limit is memory consumption due to database cache growth to support searches over both full-text and document metadata. It is possible to crawl and search over a larger corpus if you use a custom query to search only document text and not document metadata.
• Maximum number of documents per folder. The maximum recommended number of documents per folder is 200. This limitation is due to long rendering times for Web browser pages when listing the folder contents. Typical clients can require more than 10 seconds rendering the list of documents in the folder. It is possible to store up to 3,000 documents in a folder if rendering time is not a concern. This might be true if you do not expect users to browse to the contents of the folder from a Web browser, or if users are predominantly using Web folders for folder browsing.
• Maximum document size. The maximum recommended document size stored in SharePoint Portal Server is 50 MB.
• Maximum number of versions per document. The maximum recommended number of versions per document is 1,000.
• Maximum number of subscriptions per workspace. The maximum recommended number of subscriptions per workspace is 100,000. The SharePoint Portal Server subscription technology is extremely efficient. Even with 100,000 subscriptions, there is a negligible impact on server performance. Testing is continuing in this area, and it is likely that this number will increase in the future.
• Maximum number of document profiles. The maximum recommended number of document profiles is 500. The number of document profiles does not significantly affect the performance of the server.
• Maximum number of content sources. The maximum recommended number of content sources is 100.
• Maximum number of role principals per folder. The maximum number of role principals per folder can vary depending on the role. The maximum size of a security descriptor is 64 KB. This size limitation determines the maximum number of role principals. The maximum number of coordinators is 150, and the maximum number of readers is 600. Note that role principals can actually be groups. Therefore, it is possible to allow a very large number of users access to a document.
• Maximum number of categories. The maximum recommended number of categories is 500. However, this number of categories can have a negative impact on performance with certain corpus types. Key factors affecting performance are the total number of categories and the total number of documents in all categories. If the average number of documents per category is 20, then the maximum recommended number of categories is 200.

Speed of Long-Running Operations

The following operations can take a long time to complete because the operation must update information on a large number of resources:

• Changing role members on a folder affects the role memberships on all the folders that inherit from that folder. Role membership inheritance completes at a rate of approximately three folders per second.
• Copying a folder also copies all of the documents and folders that belong to that folder. The typical file-copying rate is two files per second. If the number of documents copied causes the task to take more than one hour, Microsoft Windows^® Explorer reports a time-out error. The server continues executing the move operation, and it ultimately completes successfully if there is enough hard disk space. If there is not enough hard disk space, then the operation partially completes, terminating at the point where disk space is exhausted.
• Renaming categories requires updating the categorization data on all the documents that belong to the affected categories. Categorizing documents typically completes at a rate of five documents per second. SharePoint Portal Server can rename a category hierarchy of 100 categories, with 10 documents in each category, in approximately 3 minutes 20 seconds.

Determining Maximum Throughput

Throughput testing was accomplished by progressively increasing the rate of user requests until latencies were no longer acceptable. The test server was a multi-processor 500 MHz Pentium III with 2 GB of RAM, a typical large general site configuration.

• User activity profile. The user load for all throughput testing was 95 percent common operations and five percent uncommon operations.
• Maximum number of operations per second. The maximum throughput on a quad processor 500 MHz Pentium III is 10 operations per second for access using Web folders. The maximum throughput is 1.5 operations per second for access using the Web browser. During these tests, the server is always using 100 percent of its CPU resources. The quad processor server achieved 3.7 times more throughput than an otherwise identical single processor server did for both browser and Web folder access. These results indicate that a server with more and faster CPUs should be able to achieve even greater throughput.
• Crawling. The rate of crawling can vary tremendously depending on file types and network conditions. For common document types stored on Web servers and file servers, a typical quad processor server is capable of crawling 12-18 documents per second. For example, crawling 1 million external documents at a rate of 12 documents per second would take 23 hours initially. An adaptive update can reduce crawling time by a factor of six—in this case, it would take four hours to crawl 1 million documents, assuming a typical pattern of changes and updates.
• Search. The optimal throughput of searches from the dashboard site varies depending on the number of documents included in the index. It is important to ensure sufficient RAM for the database cache to achieve optimal throughput for searches. A quad processor 500 MHz server with 2 GB of RAM and 3.5 million documents included in the index can respond to 95 percent of all searches in less than five seconds, with a maximum throughput of 20 searches per minute.
• Propagation. The amount of time required to propagate an index to the dashboard server can vary depending on the kinds of documents included in the index, and the amount of metadata in the index. Typically, propagation takes one minute for every 20,000 documents in the index.

Calculating Resource Consumption

It is very important to ensure that the server has sufficient amounts of key resources to prevent a suboptimal user experience.

CPU

The CPU resources of the server are consumed during periods of maximum usage. It is possible to ensure that there is sufficient CPU capacity on the server by following the detailed user model recommendations and maximum throughput recommendations outlined previously.

The processing required for adding documents to the index can also consume CPU resources. If the server has more than 100,000 documents in the index, it is recommended that you use a separate server dedicated to crawling content to improve the experience of users.

Hard disk

As SharePoint Portal Server stores more documents or adds more documents to the index, document storage consumes hard disk space. The following detailed guidelines help you to calculate your storage requirements depending on the particular balance of the number of documents stored within the server or available for searching:

• Base installation and workspace creation. The base installation of SharePoint Portal Server without creating a workspace consumes approximately 150 MB. The creation of the first workspace consumes an additional 50 MB. Each additional workspace consumes an additional 20 MB.
• Document and version storage. The rate of hard disk consumption varies depending on several criteria.

  Documents stored in standard folders using a document profile with 10 properties consume disk space equal to the sum of 12 KB for metadata storage, 30 percent of the document size for index storage, and the size of the document.

  Standard folder space = 12 KB + (1.3 X document size)

  Documents stored in enhanced folders using a document profile with 10 properties consume 12 KB for metadata storage for each version. In addition, index storage after the document is published accounts for 30 percent of the document size. You must also calculate 100 percent of the document size for each stored version of the document.

  If the document was never checked out (that is, the document is checked in and then immediately approved), then you can calculate the following:

  Enhanced folder space = 12 KB + (1.3 X document size)

  If the document was checked out, then the following is true:

  Enhanced folder space = (number of versions + 1) X (document size + 12 KB) + (0.3 X document size)

• Index growth. The index can grow at a widely varying rate depending on the types of documents that you add to the index. The growth rate also depends on the number of searchable words in the document. Text files and HTML files increase the size of the index faster than Microsoft Word files or Microsoft Excel files. Typical mixes of HTML files and Microsoft Office documents expand the size of the index at approximately 30 percent of the size of the original documents. The index increases in size due to documents stored on SharePoint Portal Server and documents stored elsewhere but made available for searching by use of content sources.

  A server dedicated to searching that uses an index from a server dedicated to crawling must have twice the total size of the index available as free space on the disk volume prior to the first index propagation. After the first propagation, it is necessary to have the total size of the index available as additional free space.

• Log files. SharePoint Portal Server stores all documents and metadata using database technology that supports both transactions and circular logging. The log files consume a maximum of 25 MB of hard disk space.

RAM

The optimal amount of RAM depends on the number of documents stored and the number of documents included in the index. In general, plan for a server that has base RAM of 256 MB, plus an additional 100 MB for every 100,000 documents either stored on the server or available in the index.

Backup and Restore

The total size of the backup file is the size of the data stored on the server. You can determine the size by calculating the size of the wss.mdb file and the full-text index. The rate of writing the backup file varies depending on the speed of the hard disks. The backup process does not consume a large quantity of RAM or CPU resources. A quad processor 500 MHz Pentium III server with a 40 MB per second Redundant Array of Independent Disks (RAID) 5 disk array can write the backup file at about 5 MB per second. This server can complete an online backup process for 28 GB of data in slightly over one hour.