Databases Reference
In-Depth Information
Figure 3: Fault treatment in autonomous disks
invoked.
Rule_6:
when failure(retrieve(HostID, StreamID));
if B=mapping((D=traverse_directory(StreamID)), backup);
then catch_up(D, B),
send(B, retrieve_backup(HostID, StreamID)).
The rule finds a backup disk using mapping information, invokes a catch-up
operation for the backup disk, and continues the traverse operation on the backup
disk. The flow of the fault treatment is illustrated in Figure 3. In this example,
Disk
2
tries to transmit a retrieve command to Disk
1
, but Disk
1
cannot be accessed
because of a disk failure, or software errors, in the disk's controller. Then
Rule_6
invokes a catch-up operation on the backup disk (Disk
3
), and transmits the retrieve
command to it.
2.2.4
Handling Skews
Distribution skew is basically handled by the distributed directory. The Fat-Btree
is capable of handling distribution skew. It detects the skew, and migrates overloaded
data to the logical neighboring disks. Therefore, the skew is systolically dissipated
through the cluster. Algorithms for detecting skew and migrating data are presented
in
11)
. Cluster reconfiguration after replacing a failed disk with a new disk is a
special case of skew handling for the Fat-Btree. The new empty disk is gradually
filled with migrated streams.
