Distributed File System prof. Giovanni Chiola

Migration and Repair


Sooner or later all servers will fail (with probability 1). Hence, lost stripes must be recomputed and reallocated to available servers.
A heart-beat protocol allows each server to monitor its neighbors.
If a neighbor does not provide vitality signs for a predefined period of time it is considered dead, and its stripes are considered lost.
If K functioning servers exchange their stripes, they can reconstruct lost stripes, and forward them to still available servers.
The discovery of lost stripes may require broadcast requests, as the hints on the locations of all stripes be missing.
Repair may be run as a periodic background activity. Its frequency may be increased by the discovery of neighbors death.
previous proto06.htm next