Method for improving cluster bring-up in a distributed topology liveness system
AssigneeInternational Business Machines Corporation
InventorsJohn R. Hare, Felipe Knop
Life of this patent
Estimating the term…
Abstract
A method is provided for establishing clusters in a distributed data processing environment having a plurality of nodes. The frequency of messages sent for the purpose of cluster formation is selected based on ranking of the nodes. Higher ranking nodes transmit such messages more frequently than lower ranking nodes. Lower ranked nodes thus are provided with an opportunity to join a cluster without first having to transmit their own messages to effectuate the joining.
Claims4
A method for establishing clusters in a distributed data processing environment having a plurality of nodes, said method comprising the step of:
- sending messages requesting cluster formation from a plurality of nodes to other nodes in the environment, with the frequency of said messages being sent being chosen based on node ranking, wherein higher ranking nodes transmit said messages more frequently whereby lower ranked nodes are provided with an opportunity to join a cluster without having to themselves act as a requestor for joining a cluster and wherein said ranking is determined from a node address and wherein, for an environment with N nodes, a rank, R′, of 1 is assigned to the node with highest address, and a rank of N to the node with lowest address, with other rankings being assigned in address order as a number between 1 and N and wherein the interval used between sending said messages is determined using the formula:
- K (F+(1-F)R(1/X)),
- where K is a constant;
- F is a fraction that controls how fast higher-ranking nodes send said messages;
- R is 1/R′;
- and X is a factor that is used to more drastically increase the interval between said messages.
Dependencies are inferred from the claim text.