Most Expedient Full Datacenter Restore from Cloud

The copy to cloud offload feature is used generally as a secondary backup after the initial backup on vSnap and a secondary backup to a vSnap at another site.

During the backup process, the changed blocks from the last snapshot are uploaded to the cloud as 16 MB cloud objects. These objects are first cached locally and then uploaded to the cloud several at a time in multiple threads.

During the restore process, we first mount the device on the vSnap reading only the required metadata in order to list the files on the volume. Each volume can contain one or many virtual machines or applications such as databases. Only after a request to read from the operating system to read a specific file on the volume are cloud requests made to read the actual data from the cloud. These requests are made only one at a time as the operating system only makes one read request at a time to the virtual device.

This restore process minimizes the number of requests to the cloud, which is vitally important when restoring to existing environments that have limited resources that we do not want to compromise by downloading more than is absolutely needed. For large databases, this allows for users to mount the database and run a single query without downloading the entire database from the cloud. It also allows to restore a particular virtual machine or application without downloading all the others.

This on demand approach reduces the restore speed to around 40MB/s in the best case as only one object is downloaded at a time. This limitation has never been raised as a serious issue in the past as customers are using the restore process as described above.