Performance: Increase number of GCD read queues

[djones6]

Description

Increases the number of GCD serial read queues to 4 (on OSX) or 8 (Linux, when compiled with GCD_ASYNCH).

Motivation and Context

This increases Kitura throughput by (ballpark) 5% (OSX) / 30% (Linux). The precise benefit depends on the nature of the workload (whether it is predominantly receive, send or 'think time' limited) and how many physical threads the system has.

Linux sees a universal improvement from increasing the number of queues, and in fact benefits even more from a higher number, but there are diminishing returns which are probably not worth the corresponding increase in memory footprint.

Most workloads improve on a 4-core Macbook with this change. The smaller Mac Mini shows slight regression from the change, but I would expect that 4+ cores is a more likely environment where performance is important. I do not have access to Mac hardware with >4 cores.

Linux (Ubuntu 16.04, Swift 4) - 4 cores, 8 threads

Note that the charts below show experiments with number of both GCD and Epoll queues. Only the GCD (blue) results are relevant for this PR, as I am not proposing a change for Epoll.

OSX (Sierra, Swift 4) - Mac Mini 2-core / Macbook Pro 4-core

Note that the colours mean different things to the Linux charts above: the graphs below show results on a 2-core Mac Mini (blue) and 4-core Macbook (orange).

Full data (internal link): https://ibm.ent.box.com/file/211761963959

How Has This Been Tested?

As well as the performance tests above, I have run the Kitura-net tests locally on Linux and OSX and they pass.

Checklist:

• I have submitted a CLA form
• If applicable, I have updated the documentation accordingly.
• If applicable, I have added tests to cover my changes.

[codecov-io]

Codecov Report

Merging #235 into master will decrease coverage by 0.53%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #235      +/-   ##
==========================================
- Coverage   77.72%   77.18%   -0.54%     
==========================================
  Files          47       46       -1     
  Lines        6939     6838     -101     
  Branches      568      561       -7     
==========================================
- Hits         5393     5278     -115     
+ Misses       1432     1427       -5     
- Partials      114      133      +19

Flag	Coverage Δ
#CHTTPParser	77.18% <100%> (-0.32%)	⬇️
#KituraNet	77.18% <100%> (-0.54%)	⬇️

Impacted Files	Coverage Δ
Sources/KituraNet/IncomingSocketHandler.swift	81.13% <100%> (-3.17%)	⬇️
...ces/KituraNet/Server/ServerLifecycleListener.swift	81.57% <0%> (-7.9%)	⬇️
Sources/KituraNet/HTTP/HTTPServerRequest.swift	72.41% <0%> (-4.32%)	⬇️
Sources/KituraNet/HTTPParser/URLParser.swift	97.33% <0%> (-2.67%)	⬇️
Sources/CHTTPParser/http_parser.c	55.15% <0%> (-2.4%)	⬇️
Sources/KituraNet/IncomingSocketManager.swift	85.5% <0%> (-1.45%)	⬇️
Sources/KituraNet/HTTP/HTTPServer.swift	80.35% <0%> (-1.26%)	⬇️
Sources/KituraNet/HTTPParser/HTTPParser.swift	99.02% <0%> (-0.98%)	⬇️
Sources/KituraNet/FastCGI/FastCGIServer.swift	63.58% <0%> (-0.62%)	⬇️
... and 3 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fc6e217...991efde. Read the comment docs.