[8.5.5.2 or later]

Considerations for Load Balancer IPv4 and IPv6 performance

By virtue of the requirement and design of Load Balancer for IPv4 and IPv6, performance is expected to be lower than Load Balancer for IPv4.

The key design objective of Load Balancer for IPv4 and IPv6 is to be least dependent on kernel for more efficient maintenance of the product. Kernel dependency adds on to the maintenance cost and effort both for the vendor and customers. Less kernel intrusiveness comes at a cost of lower performance. However, the performance of Load Balancer for IPv4 and IPv6 on Linux has proved to be closer to that of Load Balancer for IPv4, and therefore is the recommended platform. The order of preference of platform for performance is as follows:
  1. Linux
  2. AIX®
  3. Windows

When the managed server is on the same subnet as Load Balancer, use the MAC forwarding method. When the managed servers are on a different subnet from Load Balancer, use either encapsulation forwarding or the NAT forwarding method. The performance of encapsulated forwarding is better than Network Address Translation (NAT) because return path traffic does not pass through the load balancer. For encapsulation, configure routers to prevent return path filtering from dropping the packets.

For further performance enhancements, IBM® has a range of other products, such as Data Power® and Blade Network Switches, which are recommended, based on the requirement. Though these products might not have a one-to-one replacement of Edge Load Balancer features, it might be a good option to evaluate these products for performance, along with some of the advanced features like SSL offload, message validation, security, threat protection, response caching, message transformation, protocol conversion, SLA enforcement.

Concept topic    

Terms and conditions for information centers | Feedback

Last updated: April 10, 2014 03:11 PM EDT
File name: cmig_lb_performance.html