Capacity planning may be required if you observe one or more of the following symptoms:
The health status of the backend origin servers goes down intermittently, as observed from the varnish logs.
Origin server machines record high CPU usage.
Processing times of many requests, as recorded in the origin access logs, are unusually high.
If machines running Primetime components are appropriately configured and are equipped with the requisite hardware, these symptoms indicate an over load condition. The following observations may help you determine whether you should deploy additional origin servers or reverse proxies:
When most of the client traffic stops, fragment pushes from the packagers account for the entire load on the origin servers. If you observe overload symptoms even without client traffic, you should deploy additional origin servers to decrease the number of streams that is served from each origin.
If GET requests account for most of the load on the origin servers, beef up the varnish / reverse-proxy layer. This could mean restructuring the varnish layer into two tiers, each tier comprising a set of servers that perform a specific function. The first tier interacts directly with the origins. The second tier, which comprises a set of reverse proxies, receives data from the first tier. This arrangement can effectively decrease the fan-out between the first reverse-proxy layer and the origins.
You can utilize auto scaling if you deploy the reverse proxies in the cloud to handle large variations in the client traffic.