Incomplete test cases - compairing apples and orranges about frameworkbenchmarks HOT 2 CLOSED

While I agree on grounds of purity, I don't think adding trivial synthetic routing to the platforms that do not provide native routing (e.g., Netty) is going to affect the numbers to a measurable degree.

To add synthetic routing to Netty, we would either add a HashMap as a lookup table or simple String comparison as we did in the JavaScript tests. E.g.,

if ("/json".equals(uri))

In extremely rough spot checking on our i7 hardware:

• Java performs 1B lookups on a 1-entry HashMap in ~6100ms for ~6.1ns per lookup. The added expense of 100,000 HashMap lookups (one lookup per request) would be roughly 0.61ms in total.
• We can execute 1B trivial Java String comparisons in ~2700ms for ~2.7ns per String comparison. We could expect that roughly 0.27ms would added for a full test of 100,000 requests.

Bear in mind that the rough numbers above were gathered in a single-threaded test. Total additional time would probably be even lower with concurrency.

I agree on the purity goal though, so we should do this change.

from frameworkbenchmarks.

Closing - summarized a bit in documentation.

from frameworkbenchmarks.