Abstract: A map-reduce compatible distributed file system that consists of successive component layers that each provide the basis on which the next layer is built provides transactional read-write-update semantics with file chunk replication and huge file-create rates. Containers provide the fundamental basis for data replication, relocation, and transactional updates. A container location database allows containers to be found among all file servers, as well as defining precedence among replicas of containers to organize transactional updates of container contents. Volumes facilitate control of data placement, creation of snapshots and mirrors, and retention of a variety of control and policy information. Also addressed is the use of distributed transactions in a map-reduce system; the use of local and distributed snapshots; replication, including techniques for reconciling the divergence of replicated data after a crash; and mirroring.

Abstract: A key-value store provides column-oriented access to data in a distributed and fault tolerant manner. Data can be inserted into the data store and data can be retrieved either randomly or sequentially from the data store at high rates. Keys for a table are ordered and the entire table is divided into key ranges. Each key range is handled by a table which itself is divided into key ranges called a partition. Partitions are also divided into segments. Such recursive division into smaller and smaller key ranges provides parallelism. At the highest level, operations on tablets can be distributed to different nodes. At lower levels, different threads can handle operations on individual segments. Large-scale restructuring operations can be decomposed into operations on individual segments so that a global lock on larger objects does not need to be kept across the entire operation.

Abstract: A map-reduce compatible distributed file system that consists of successive component layers that each provide the basis on which the next layer is built provides transactional read-write-update semantics with file chunk replication and huge file-create rates. Containers provide the fundamental basis for data replication, relocation, and transactional updates. A container location database allows containers to be found among all file servers, as well as defining precedence among replicas of containers to organize transactional updates of container contents. Volumes facilitate control of data placement, creation of snapshots and mirrors, and retention of a variety of control and policy information. Also addressed is the use of distributed transactions in a map-reduce system; the use of local and distributed snapshots; replication, including techniques for reconciling the divergence of replicated data after a crash; and mirroring.

Abstract: A map-reduce compatible distributed file system that consists of successive component layers that each provide the basis on which the next layer is built provides transactional read-write-update semantics with file chunk replication and huge file-create rates. A primitive storage layer (storage pools) knits together raw block stores and provides a storage mechanism for containers and transaction logs. Storage pools are manipulated by individual file servers. Containers provide the fundamental basis for data replication, relocation, and transactional updates. A container location database allows containers to be found among all file servers, as well as defining precedence among replicas of containers to organize transactional updates of container contents. Volumes facilitate control of data placement, creation of snapshots and mirrors, and retention of a variety of control and policy information.

Abstract: A system and method for accelerating file deletion by preloading indirect blocks. When processing level 1 (L1) indirect blocks, the file server issues a read request for the next N indirect blocks to be loaded into a buffer cache. The file server processes the indirect blocks from the buffer cache without requiring multiple single block disk accesses. Once the N indirect blocks have been processed, a read request for the next N L1 indirect blocks is then sent. The technique permits parallel processing by loading indirect blocks to be processed while previously cached indirect blocks are being processed.