Abstract: When a backup client sends a request to back up a file to a backup server, the file and an index (e.g., checksum, hash, encryption, etc.) of the file are stored on the backup server in an efficient deduplication storage. If a backup client sends a request to back up a modified version of a file already stored on a backup server, the modified portion of the file is stored. In addition, an index of the modified portion is generated and stored along with the modified portions on the backup server. The indices can be used to reconstruct the file or modified version of the file when retrieved. The efficient deduplication storage method ensures that multiple copies of files or portions of files do not exist on the servers.

Abstract: Data from computing devices is backed-up regularly, storing a snapshot of the data and corresponding metadata in a data store. The backup data are stored for a relatively short period of time before being archived to long-term storage. For snapshots with files with identical data that are not archived together, archive storage space and computing resources may be conserved by not storing duplicates of the data. When the data is added to the archive storage, the archive storage location is added to backup reference entries for other files with identical data. When all files referencing an archive storage location are expired from the backup data store, an archive retention period is initiated, and an entry is added to a time-based archive expiration database indicating the storage location and an expiration time for the archived data. At the expiration time, the archived data is designated for deletion from the archive.

Abstract: Disclosed embodiments include a method (system and non-transitory computer storage readable medium) for load-balancing a distributed database. The distributed database includes one or more storage machines configured to store a plurality of partitions, where each partition includes key-value pairs. In one embodiment, the distributed database prepares for load-balancing by determining a partition to redistribute (or repartition) and generating smaller partitions of the determined partition. In one aspect, each of the smaller partitions is smaller than the determined partition. The redistribution of the partition can occur, when an amount of requests to access one or more key-value pairs stored in the database increases beyond a predetermined request level or when the size of a partition exceeds a predetermined size. Key-value pairs of the determined partition can be split into different sets of key-value pairs, and each set of key-value pair is copied to a corresponding smaller partition.

Abstract: A system and a method are disclosed for pre-seeding data to backup servers and determining servers for additional backups. Backup data is received from a client device through a first backup request and sent to a primary server for storage. Additional backup data from a second backup request is received. After the initial backup data is stored on the primary server and pre-seeded by the primary server on a secondary server, a status of backup servers associated with the client device is received. The backup servers include the primary server and can include the secondary server. Responsive to the status indicating availability of a server in the backup servers, a recipient server is identified from the backup servers and the additional backup data is sent to the identified recipient server.