Utilizing a load balancer with ECS is highly recommended. Highlights some of the best practices when deploying with ECS include:
- Do not use an LTM for CAS traffic. The Centera SDK has a built-in load balancer and cannot function without direct access to all nodes.
- Traffic management is best utilized for data traffic.
- Terminate SSL connections on LTM when possible to reduce the load on the ECS nodes.
- If SSL termination is required on ECS nodes, use Layer 4 (TCP) to pass through the SSL traffic to ECS nodes for handling. The certificates here are installed on the ECS nodes and not on the LTM.
- Use two or more LTMs at each site. Using a single LTM at a site creates a single point of failure. In multisite federated environments where BIG-IP DNS devices provide site redundancy, a single LTM may be sufficient so long as applications can tolerate increased latency during a site failure. If application traffic cannot be directed to alternate sites during failure, a single LTM deployment can result in complete data unavailability.
- For NFS, use only the HA functionality of the load balancer. That is, for NFS workloads configure LTM to keep client sessions terminated on a single ECS node. Only during node failure should an NFS session be torn down and established on another ECS node. Balancing NFS traffic across ECS nodes is inefficient because it doesn't take advantage of ECS caching.
- When deploying three or more ECS sites, and when performance considerations allow, employ a global load balancing mechanism to distribute load across sites to take advantage of ECS XOR storage efficiency. It is also important to optimize the local object read hit rate in a global deployment.
- For sites serving a large amount of traffic from a small number of end-points, using source-address based session persistence may result in an uneven distribution of load. If this is an issue, restrict session persistence to only "PUT" and "POST" operations, if possible, or consider another session persistence profile for the workflow.
- Service providers and others accepting connections from the public Internet should use the "TCP WAN Optimized" profile. This enables TCP window scaling and other RFC 1323 throughput- improving options, adjusts various buffer sizes upward, and sets other defaults that are appropriate for WAN traffic. Service providers should also consider raising TCP send and receive buffer sizes and proxy buffer high and low water marks even further to at least 128KB or 256KB. This allows a better throughput from end users on links with high bandwidth delay factors (for example, high bandwidth and high latency).