Current performance (latency) of different network-transport backends

results0.1

This is the average latency as measured by a single trial run of 100K ping/pongs with several different transports.

    PingTCP:           7.34s   >80%CPU 
    PingTCPTransport:  19s     ~60%CPU
    PingPipes:         6.36s   >=100%CPU

The machine is a 3.1 ghz Intel Westmere (Dell workstation) running RHEL 6.2.

The PingTCP is a baseline that uses the raw Network.Socket interface rather than the Network.Transport abstraction. 

I just ran on granite.cs.indiana.edu with the following (discouraging) results:

afoltzer@granite ~/src/distributed-process [master*]$ ./benchmarks/PingTCP client granite.cs.indiana.edu 8190 100000
server: listening for pings
warming up
estimating clock resolution...
mean is 1.579211 us (320001 iterations)
found 2315 outliers among 319999 samples (0.7%)
  2202 (0.7%) high severe
estimating cost of a clock call...
mean is 45.47631 ns (11 iterations)
found 2 outliers among 11 samples (18.2%)
  1 (9.1%) high mild
  1 (9.1%) high severe

benchmarking PingTCP
collecting 100 samples, 1 iterations each, in estimated 349.7981 s
mean: 3.355803 s, lb 3.274801 s, ub 3.419473 s, ci 0.950
std dev: 367.1117 ms, lb 298.1148 ms, ub 450.5691 ms, ci 0.950
found 9 outliers among 100 samples (9.0%)
  5 (5.0%) low severe
  4 (4.0%) low mild
variance introduced by outliers: 82.101%
variance is severely inflated by outliers

afoltzer@granite ~/src/distributed-process [master*]$ ./benchmarks/PingTransport client granite.cs.indiana.edu 8191 sourceAddr 100000
warming up
server: awaiting pings
estimating clock resolution...
mean is 1.396276 us (640001 iterations)
found 3593 outliers among 639999 samples (0.6%)
  3548 (0.6%) high severe
estimating cost of a clock call...
mean is 43.10540 ns (11 iterations)

benchmarking PingTransport
collecting 100 samples, 1 iterations each, in estimated 826.5324 s
mean: 8.009859 s, lb 7.816015 s, ub 8.107999 s, ci 0.950
std dev: 681.3909 ms, lb 398.1707 ms, ub 1.071390 s, ci 0.950
found 13 outliers among 100 samples (13.0%)
  10 (10.0%) low severe
  3 (3.0%) high severe
variance introduced by outliers: 73.785%
variance is severely inflated by outliers

Well those results are quite a bit better than mine (from a seemingly identical machine!).

You got 3.35s instead of 7.3 (TCP)
And 8.0s instead of 19 (Network.Transport.TCP)

What gives? I'll check if it's the problem you mentioned with certain port ranges.