fphttp slower with threads

[Trenatos]

I'm playing with fphttp to create an API and comparing it to my usual stack of OpenBD (CFML engine running on the JVM), I also have a simple Java Spring API to compare with.

I'm comparing them locally with WRK2 using

Code: Pascal  [Select][+][-]

1. wrk2 -t2 -c100 -d30s -R2000 --latency http://localhost:7080/status/

In the process I've done both x86 and x64 with and without threads and had some odd results.

Both architectures have about 13% less throughput when setting multithreading to true, and the x64 is a tad slower than the x86 version, but only by a little.

Compared to my usual stack both have about 26% less throughput.

I'm surprised that they're slower than the Java versions at all, that's a big disappointment as I was hoping for better performance with a native binary solution compared to Java, and I'm surprised that threading caused a drop in throughput.

I ran each test a good handful of times and the average I got was:
OpenBD - 1995 requests/sec
Java Spring API - 1997 requests/sec

FPC fphttp
No threading
x86 - 1560 requests/sec
x64 - 1450 requests/sec

FPC fphttp
With threading
x86 - 1480 requests/sec
x64 - 1370 requests/sec

Tested on my Macbook Pro.

[marcov]

What does the Mac Brook  pro run?   Afaik OS X has several threading APIs,  (POSIX and native?), maybe it is merely a result of which API you pick.

[Trenatos]

It's running OSX El Capitan

System report:

Model Name: MacBook Pro
Model Identifier: MacBookPro11,5
Processor Name: Intel Core i7
Processor Speed: 2.5 GHz
Number of Processors: 1
Total Number of Cores: 4
L2 Cache (per Core): 256 KB
L3 Cache: 6 MB
Memory: 16 GB

I'm using these options in my fpc app

Code: Pascal  [Select][+][-]

1. {$ifdef darwin}
2. cthreads,
3. cmem,
4. {$endif}

[Leledumbo]

It actually also happens on Linux, using cthreads, with both tested and tester programs running on the same machine. I have yet to compare using server to server setup, to avoid any intra resource sharing problem, but I suppose fpc would shine in such a setup.

[Trenatos]

I might give it a go under more proper server settings (Separate server instance from test machine, local network to avoid latency issues/spikes, etc.)

I'll come back with results of my tests when I get a setup going.