SharePoint Online Data Resiliency
A key principle for SharePoint Online is to never have a single copy of any piece of data. SharePoint Online uses SQL Server replication, which is a set of technologies for copying and distributing data and database objects from one database to another, and then synchronizing between databases to maintain consistency.
For example, when a user saves a file in SharePoint Online, the file is chunked, encrypted, and stored within Azure Blob storage. Azure Blob service provides mechanisms to ensure data integrity both at the application and transport layers. This post will detail these mechanisms from the service and client perspective. MD5 checking is optional on both PUT and GET operations; however, it does provide a convenience facility to ensure data integrity across the network when using HTTP. Additionally, since HTTPS provides transport layer security additional MD5 checking is not needed while connecting over HTTPS as it would be redundant. Azure Blob service provides a durable storage medium, and uses its own integrity checking for stored data. The MD5's that are used when interacting with an application are provided for checking the integrity of the data when transferring that data between the application and service via HTTP.
To ensure data integrity the Azure Blob service uses MD5 hashes of the data in a couple different manners. It is important to understand how these values are calculated, transmitted, stored, and eventually enforced to appropriately design your application to utilize them to provide data integrity. For more information, see Windows Azure Blob MD5 Overview.
Metadata and pointers to the file are stored in a SQL Server database (the content database). All the chunks – files, pieces of files, and update deltas – are stored as blobs in Azure storage that are randomly distributed across multiple Azure storage accounts. The SQL database is hosted on a RAID 10 storage array which is synchronously mirrored to another RAID 10 storage array in a separate rack within the same datacenter. Asynchronous log shipping is then used to replicate the data to another RAID 10 storage array in a second datacenter. In addition to protecting data with RAID 10 and synchronous and asynchronous replication, scheduled data backups are taken which are also asynchronously replicated to the second datacenter.
In SharePoint Online, data backups are performed every 12 hours and retained for 14 days. SharePoint Online also uses a hot standby system that includes paired geographically-separate datacenters within the same customer data location region (for example, Chicago and San Antonio for customers who have provisioned their tenant in the United States) configured as active/active. For example, there are live users that have Chicago as their primary datacenter and San Antonio as a failover datacenter, and live users that have San Antonio as their primary datacenter and Chicago as their failover datacenter.