The challenge of these problems is that there is no central control node.
Source nodes can only disseminate data to the networks from their local information,
and querying nodes have to forward requests to nodes that can
provide them with desired data or initiate detection as soon as possible. Quorum
methods are designed to facilitate these operations. A subset of nodes
is collected to form a write quorum, in which update operations take place.
254
A potential di?®erent subset of nodes is collected to form a read quorum, so
that retrieval processes are executed. The availability of this quorum method
lies in that each quorum proceed with enough nodes to make sure intersection
between write and read quorums exists.
Traditional protocols solving these problems prepare each node to participate
in data accessing, either in forming write quorums or read quorums.
The system overhead is O(n). Later on, networks can be divided into several
hierarchies. One or more nodes in each level are selected to be members of
each quorum. Thus the size of quorum becomes proportional to the height of
the hierarchy. Another significant improvement in reducing system overhead is
that by conceptually and evenly dividing the networks into rows and columns,
the size of quorums has decreased to O(???n). Moreover, negotiation and frequency
control can be combined in constructing quorum systems. Negotiation
helps avoid redundant data dissemination.
Pages:
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416