-
Notifications
You must be signed in to change notification settings - Fork 3
Conceptual architecture and main terms
Consider a distributed Bob cluster that consists of nodes, each storing some part of data. Each node, in turn, uses a set of physical disks on which it stores a set of VDisks. VDisks are replicated between nodes. Bob node chooses VDisk to store data using distribution function (currently it is mod).
The whole dataset is logically partitioned into a predefined number of VDisks, each assigned a unique number ranging from 0 to N-1, where N is the total number of VDisks. The number of VDisks is specifically chosen to be several times larger than the number of physical disks in the cluster (also taking into account potential future cluster expansion).
Pearl is append only key-value blob storage on disk. Used in bob as backend.
Blob (Binary large object) is a collection of binary data stored as a single entity.
Node running instance of bob. Usually dedicated server in cluster.
Cluster is set of nodes that work together and presented in the same config file.
VDisk is virtual storage that stores a unique subset of data from a dataset. Vdisk has a unique identifier and replicated across nodes of cluster.
Alien is a folder for the data that cannot be written to its node due to failure. It has the same structure as the normal folder but also has "node name" in folder hierarchy.