More tests?

[halgari]

Can we get tests from other boxes? On my machine (4 core hyper-threaded OSX MBP circa 2012):

lein core.async:
Evaluation count :
120 in 60 samples of 2 calls.
Execution time mean : 615.162515 ms
Execution time std-deviation : 8.065130 ms
Execution time lower quantile : 606.073498 ms ( 2.5%)
Execution time upper quantile : 633.294586 ms (97.5%)
Overhead used : 2.123800 ns

lein pulsar:
Evaluation count : 120 in 60 samples of 2 calls.
Execution time mean : 631.772231 ms
Execution time std-deviation : 24.939520 ms
Execution time lower quantile : 582.649998 ms ( 2.5%)
Execution time upper quantile : 681.948148 ms (97.5%)
Overhead used : 2.126756 ns

[pmbauer]

Sure.
And here's another data point, another machine.

$ sysinfo

============= CPUs =============
model name  : Intel(R) Core(TM) i7 CPU         980  @ 3.33GHz (x 12 cores)
============= MEM =============
24128 MiB
...

$ uname -a
Linux capsaicin 3.13.0-14-generic #34-Ubuntu SMP Thu Feb 27 22:32:54 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux

$ java -version
java version "1.7.0_51"
Java(TM) SE Runtime Environment (build 1.7.0_51-b13)
Java HotSpot(TM) 64-Bit Server VM (build 24.51-b03, mixed mode)

$ lein core.async
(thread-ring 503 500000) => 19
[...]
Evaluation count : 60 in 60 samples of 1 calls.
             Execution time mean : 2.626226 sec
    Execution time std-deviation : 157.820088 ms
   Execution time lower quantile : 2.177659 sec ( 2.5%)
   Execution time upper quantile : 2.739989 sec (97.5%)
                   Overhead used : 2.517213 ns

Found 5 outliers in 60 samples (8.3333 %)
    low-severe   4 (6.6667 %)
    low-mild     1 (1.6667 %)
 Variance from outliers : 45.0962 % Variance is moderately inflated by outlier

$ lein pulsar
(thread-ring 503 500000) => 19
[...]
Evaluation count : 120 in 60 samples of 2 calls.
             Execution time mean : 804.934861 ms
    Execution time std-deviation : 13.916416 ms
   Execution time lower quantile : 762.460052 ms ( 2.5%)
   Execution time upper quantile : 816.690026 ms (97.5%)
                   Overhead used : 2.522869 ns

Found 9 outliers in 60 samples (15.0000 %)
    low-severe   6 (10.0000 %)
    low-mild     3 (5.0000 %)
 Variance from outliers : 6.2795 % Variance is slightly inflated by outliers

[pmbauer]

I'm consistently seeing pulsar take full advantage of multiple cpus with this benchmark while core.async does not.
Note the environment stats I posted and that this is a new machine installation.
The only obvious difference I see between our environments is you are on a Mac.

[pmbauer]

P.S. @halgari
I hadn't seen your tweet earlier.
Of course no one goes around passing things in a ring for fun; this is just the classic erlang thread-ring benchmark and it measures communication overhead (pardon if you already know this).
Please forgive any perceived slight against core.async or your work; I use it and love it.

I am wondering if there is something specific to Linux/JVM that shows core.async poorly at least for this one contrived case.

[halgari]

No slight taken. This is a benchmark I've performed in the past, and it's good to see other datapoints. I know Pulsar uses FJ pools, so I'll dig into seeing how hard it would be to use those in core.async, and what difference it makes.

[halgari]

Also if I'm reading the code right, this is only sending a single message through the ring. This would mean that at most only 2 CPUs really should be in use, a one would be sending the int and the other would be receiving. All the rest should be blocked.

[pmbauer]

correct