NormalizeHost CPU Performance

[murphymj25]

In looking at a CPU pprof of one of our production instances, we have noticed that a significant amount of CPU time is being spend on the strings.ToLower in the normalizeHost function in table.go.

The pprof was taken on a 16 core instance, with about 3,000 requests per second of traffic. In this environment, the route table has about 13,000 separate routes. In the pprof, we see the following:

I've done some testing and updated the matchingHostNoGlob to use string.EqualFold instead of doing the comparison with the normalizeHost and strings.ToLower. With that change, the normalizeHost was reduced from 22% to 8% with similar request volume and route table size. The overall CPU usage of the instance was reduced by about 10%.

I'll create a PR to share the changes I've done already. I think there's still some opportunity to further reduce the CPU load by changing how the :port is trimmed off.