Browse Prior Art Database

[WebSphere] System and Method for dynamic self balancer leader election

IP.com Disclosure Number: IPCOM000227732D
Publication Date: 2013-May-14
Document File: 4 page(s) / 366K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a new mechanism for implementing self-balancer cluster. Different from traditional self-balancer design, this disclosure allows a self-balancer cluster to dynamically assign a new leader of the balancer group based on the performance of each member in the group. Therefore, the balancer group can be more efficient and balanced than traditional design does.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 4

[WebSphere] System and Method for dynamic self balancer leader election

For large scale network infrastructure, load balancer is a widely used network configuration to distribute the incoming traffic equally to backend servers. However, the load balancer nowadays still can not totally get rid of the bottleneck and single point of failure problems. This is because in order to setup the load balancer, people usually place device(s) in front of the backend servers to distribute the traffic, like following figure 1.

     Self balancer is another mechanism to solve the traffic loading problem. Within a the self balancer group, there will be a leader to take the responsibility to distribute the incoming traffic. Thus, there will be no extra devices needed to distribute the traffic and the traffic can be load balanced by the group of machines. However, the problem of self balancing is the loading of the leader will increase. If the processing loading of leader is high then the overall performance will be affect by slow load balancing that the leader can provide.

1


Page 02 of 4

     The core idea of this disclosure is dynamically assign the leader of the self balancer group by evaluating various performance related metric of each appliance. Since each appliance in the self-balancer group will periodically exchange their status ( mostly the availability) using HSRP "Hello" message, the internal performance metric could be used to infer the relative "priority value" to be used in the HSRP Hello message as an approach for distributed status exchange. This allow appliance with more computing resource to preemp the leader role from existing leader and therefore balance the load within the cluster.

     Besides, to prevent the situation where the HSRP preemption occur excessively within the local network, an asymmetric priority scheme is introduced. The leader will use a linear priority level so that its internal status could be easily detected by other appliance, while the other appliances in the network will use a segmented priority level to avoid unnecessary contention. More details could be found in the section 3.

The summaries of this disclosure could be listed as below:


1) Dynamically assign self balancer leader in the cluster using HSRP protocol


2) A dynamic HSRP "PRIORITY" value is calculated based on the performance metric of each appliance

3) Using an asymmetric PRIORITY scheme to avoid excessively HSRP preemption in the network. Different from the traditional self-balancer design, the disclosure dynamically assigns a

normal working appliance, called leader, in the self-balancer group for receiving and handling the inbound requests, as shown in figure 3.

2


Page 03 of 4

The design details are summarized as below:


a. For dynamically assign the leader

     1) Each appliance in the group needs to calculate a priority value by evaluating related metrics of its performance, so appliance with high system loading will have low priority and vi...