SharePoint Server 2013 WAN case studies: United Airlines, Teck, and Fabrikam
Summary: Learn how three global companies designed their SharePoint architectures to meet their requirements.
Applies to: SharePoint Server 2013
In this article:
United Airlines, Fabrikam, and Teck worked with Microsoft in the prerelease program for SharePoint Server 2013. The IT teams for these companies were led by the following individuals.
Denise Wilson — Senior Manager of Enterprise Collaboration Services, United Airlines
Mahmood Jaffer — IT Specialist and SharePoint Architect for Teck, Softlanding (www.softlanding.ca)
Fabrikam is a fictionally named company representing a large world-wide manufacturing company. While the Fabrikam IT team remains anonymous for this case study, the team was very active in providing valuable feedback to Microsoft.
Microsoft Senior Consultant, Faizan Khan, worked with these companies to evaluate WAN performance and recommend appropriate architectures using SharePoint 2013.
“Getting a chance to work on the latest version of SharePoint with three global customers with unique scenarios was rewarding. I enjoyed tackling the previous SharePoint 2007 and 2010 paradigms related to architecture and WAN connected users head on with all of the new SharePoint 2013 capabilities. It was great to see how SharePoint 2013’s higher scalability and optimizations for connectivity over the WAN allowed each customer to reduce their server footprint while providing an excellent social collaboration experience for their own customers. When you start planning your SharePoint 2013 architecture I highly recommend taking a look at the new opportunities to consolidate your infrastructure as these customers have done.” –Faizan Khan
United Airlines—Central deployment, single datacenter
United Airlines is one of the world’s leading airlines serving 379 global airports on six continents with over 85,000 employees. United Airlines participated in the SharePoint Server 2013 Rapid Deployment Program and the Windows 8 and Windows Server 2012 Technology Adoption Programs.
Denise Wilson, Senior Manager of Enterprise Collaboration Services team at United Airlines, worked with Microsoft on several projects with SharePoint Server 2013. Wilson worked with Microsoft Consulting Services to architect the new SharePoint Server 2013 environment and to provide feedback about the airline’s experience using SharePoint Server 2013 in its WAN environment. Wilson also worked closely with the SharePoint product team to develop and pilot a new procurement tool that is based on taxonomy and new search features in SharePoint Server 2013.
United Airlines’ employees rely on SharePoint sites to complete job-related tasks at every company site, including the 379 global airports. The route map for the airline shows the geographic environment. The whole environment is served by a centralized SharePoint farm.
The environment includes high latency connections and other challenges that relate to performance. Wilson’s team addressed these challenges in many ways. The following table summarizes some performance optimizations that made a positive difference for the company. The third column of the table includes additional options that other organizations with similar challenges can consider.
|Challenges||United optimizations||Additional options that other organizations can consider|
High utilization, low-bandwidth connections at international locations
Quality of service (QoS), browser compression (Internet Explorer)
Performance at airport kiosks
Prioritizing customer kiosk performance over back office performance of SharePoint traffic
QoS, browser compression, local caching
SQL Server Master Data Services (MDS), network interface card optimization
Intranet SharePoint performance
Browser compression, Internet Information Services (IIS) File Compression (static and dynamic), binary large object (BLOB) Caching, Output Caching, Object Caching
SQL Server performance optimization
Memory tuning, reducing databases, reorganizing indexes, and rebalancing the virtual log files
Do not enable auto-create statistics
Set max degree of parallelism (MAXDOP) to 1
The last row of the table, SQL Server performance optimization, highlights the importance of paying attention to the overall health and performance of a SharePoint farm. While these are not WAN-specific optimizations, the United Airlines team reported noticeably faster SharePoint performance across the environment after optimizing compression and caching.
United Airlines operates its SharePoint environment in a Chicago-based datacenter. The disaster recovery environment is hosted in its Houston datacenter. Currently the company uses log-shipping between farms. Failover between farms is scripted and can be achieved in about 15 minutes. The following diagram shows the disaster recovery environment.
“There are a number of options for supporting high availability and failover,” states Khan. “This example, log shipping between farms across the WAN, is fairly standard. United Airlines has tiered their services and databases. Some databases have a high or medium SLA and are log-shipped. Others, such as content databases, have a lower SLA and are backed up and then restored. While log-shipping is a common way to build a failover solution, database mirroring can also be used, as well as high availability groups with SQL Server 2012.”
The overall performance in SharePoint Server 2013 is optimized for building and operating larger farms. Even with the high number of users, large volume of content, and extensive use of the feature set, the Microsoft Consulting team and the SharePoint product team feel comfortable recommending a single, large farm for United Airlines.
The farm featured in the Enterprise-scale model resembles the farm designed by Khan for United Airlines. This model provides guidance for patching and updating farms of this size.
Fabrikam—Central deployment, multiple datacenter
Fabrikam is a fictional company that represents a large world-wide manufacturing company that participated in the SharePoint Server 2013 prerelease program. Fabrikam has several-hundred-thousand employees and many content farms that were migrated from the previous version.
A priority for Fabrikam is offering My Sites to all employees. With the large number of employees, much consideration went into the architecture for My Sites. Fabrikam weighed the benefits and risks of deploying one My Site farm versus multiple My Site farms. The following table summarizes the advantages and disadvantages for both architectures.
|Architecture 1 — One large My Site farm and multiple content farms||Architecture 2 — Two or three My Site farms and multiple content farms|
All employees are served by one My Site farm. The My Site farm and the content farms are located in the same datacenter and consume the User Profile service from a dedicated services farm which is also located in the same datacenter.
Two or more My Site farms are deployed and employees are spread across the My Site farms. Each My Site farm consumes a dedicated User Profile service application. The My Site farms and the User Profile service applications can be spread across datacenters if the User Profile service application that is associated with a My Site farm is located in the same datacenter. Content farms can be spread across datacenters. However, each content farm must consume a User Profile service application that is located in the same datacenter.
Single User Profile service to manage.
Single My Site farm to manage.
With a single User Profile service application, all social features are expected to work as expected across farms.
My Site farms are smaller.
The environment can be distributed across multiple datacenters.
Users can be assigned to My Site farms based on location, which provides greater opportunities for maintenance windows.
Concern about outgrowing a single My Site farm.
With a world-wide user base, scheduling maintenance windows is more challenging.
More farms to maintain.
With multiple User Profile service applications, social features across farms are limited.
An advantage of the performance improvements in SharePoint Server 2013 is the ability to combine content and users into a smaller number of large farms. These improvements helped Fabrikam to feel comfortable with a single My Site farm (Architecture 1 in the previous table). This architecture avoids the complexity of managing multiple User Profile service applications. However, with a single User Profile service application, all of Fabrikam’s content farms must be located in the same datacenter. The following diagram shows Fabrikam’s environment.
In the diagram, a separate datacenter hosts Fabrikam’s search farm and a dedicated search team manages the farm.
Khan explains the design challenge: “Because the social capabilities of SharePoint Server 2013 are extremely important to Fabrikam, we engaged in a number of detailed conversations on the best approach to implementing a My Sites farm. Should we implement one large farm or fewer smaller farms? There are pros and cons with either choice including manageability, availability, and functionality. There are also concerns with scalability both on the application and storage layers. With new guidance from the SharePoint product group, we decided the best approach is to implement a single farm. This approach results in the best social experience while remaining within the scalability guidelines of SharePoint Server 2013. Due to the large user base, Fabrikam highlights the ability of SharePoint Server 2013 to scale while minimizing the number of farms required.”
For more information about how to design architectures that have multiple farms and multiple datacenters, see the following model: Multi-farm architectures with SharePoint Server 2013.
A goal of working with Fabrikam in the prerelease program was to understand how SharePoint Server 2013 performs over a variety of WAN connections. Fabrikam used Visual Studio to script a load test that is made up of many unit tests and then ran the load test from multiple geographic locations. For more information about how to create a similar test, see SharePoint 2013 WAN testing with Visual Studio 2012 walkthrough.
In this first set of results, two users in the Fabrikam Shanghai, China, office ran the load test against the SharePoint sites in the Texas, USA, datacenter. Latency is about 190 milliseconds roundtrip. The upload, download, and Outlook Web Access (OWA) tests were conducted with a 1 mb file.
Fabrikam — WAN performance across the feature set for Shanghai to Texas
The test results show that performance is good, especially for the social tasks.
The next set of results shows performance for the same load test across a larger set of geographic locations where Fabrikam employees work. The servers running SharePoint Server 2013 are located in Texas, USA.
Fabrikam— Results across the feature set for different locations
Even though there are varying degrees of latency, performance is good for users across the globe. The Fabrikam test results provide an example of systematic WAN testing by using a load test that is comprised of a variety of SharePoint tasks that are important to the company.
Fabrikam is an example of a world-wide company that is succeeding with a central datacenter model, instead of deploying SharePoint Server 2013 to multiple regions across the world. If you are considering a move from a central datacenter model to multiple SharePoint sites in different geographical regions, make sure that you conduct WAN testing to see whether it is really necessary.
Teck—geographically dispersed deployment
Teck is a mining company with many remote sites throughout the world. Many sites are connected to the Teck WAN by high latency and intermittent satellite connections. Mahmood Jaffer, IT Specialist and SharePoint Architect with Microsoft Partner Softlanding, worked with Teck to design and support their global SharePoint architecture.
The description of the Teck architecture includes several illustrations that might be difficult to read within this article. These illustrations are also available in SharePoint 2013 WAN architecture diagrams.
Teck uses two datacenters in Calgary, Canada, and Santiago, Chile. The following map shows the two datacenters plus the 10 remote sites.
Even though some remote sites are located on the same continent as a datacenter, these sites rely on low bandwidth connections or satellite connections that are intermittent. For sites shown on the map, WAN performance prevents or discourages users from uploading content to a SharePoint site at one of the two datacenters, or a site located at a nearby site.
Teck weighed the costs and benefits of deploying many regional SharePoint farms compared to not using SharePoint sites. For the mining company, the benefit of using SharePoint Server 2013 features to manage content outweighs the administrative costs of deploying many farms to support the distributed environment. In particular, the company relies on the following capabilities:
Managed metadata—a standard tagging scheme that is applied to content so that assets can be discovered and managed appropriately over time.
Enterprise-wide search—the ability to discover content assets that meet specific criteria, which metadata defines, and to know the mining sites that these assets are associated with.
Integrating the environment to support these features requires coordinating service applications at all farms. The User Profile Service and My Sites are deployed to each farm.
With this highly distributed environment, Teck is considering using Yammer for social features.
Search is an important component of the enterprise content management strategy of Teck. The company searches for assets that meet specific criteria. The company also requires the ability to easily determine the mining sites that assets are associated with. Currently much of the historic data that belongs to Teck resides on file shares. It will take some time for the company to move the content to a SharePoint environment. Some assets might remain on file shares indefinitely.
At the remote sites, search is deployed to each local farm and crawl rules are set up to also crawl local file shares. The following diagram shows the configuration that is replicated for all remote sites.
Enterprise-wide search is provided at the company’s two datacenters in Calgary and Santiago. Teck considered using a combination of federated search and remote result sources to display results. Federated search was considered for sites with intermittent satellite connections. By configuring federated search to show results in a separate Web Part Teck would have a visual indication of whether the remote site is connected or disconnected at the time of the search. However, Teck decided to use remote result sources for all sites to ensure that all results are security trimmed.
The following diagram shows how enterprise-wide search is architected to include results from all remote sites.
The relationship of the two datacenters to each other is not included in the illustration. Each datacenter uses remote result sources to include the other datacenter in search results.
Finally, the Managed Metadata service application is integrated by deploying it to the central farm in Calgary and then sharing this service application to all other farms in the environment. The Managed Metadata service application is one of the few service applications that can be shared across farms that are separated by WAN links.
One issue that Teck reports with this configuration, however, is that the Managed Metadata fields and tagging are not available to remote farms when a satellite connection is disconnected.
Khan summarizes the design challenges for Teck: “Because Teck supports the collaboration needs of occasionally connected offices or offices with extreme latency due to the use of satellite links, we needed to build an architecture that was flexible while minimizing the complexity as much as possible. The WAN improvements helped Teck consolidate some of the remote farms, but not all. The result is two central sites that act as the hub for a number of smaller farms. This provides a cohesive enterprise-wide search experience while supporting the goal of providing remote office users with access to SharePoint sites even when disconnected from the ‘hub’.”