Improved sync/async benchmarks

[segabriel]

No description provided.

[artem-v]

Better just propagate with Exceptions.propagate(e) . Can u put right after this comment an improvement whihch this parallel thing is giving to us. I heard u told it's like 2x perf boost. Ok, can u put the stats right here. Like, here it was, here it became.

[artem-v]

Cool!

[artem-v]

As you have said runForAsync doesn't work well with this parallel() runOn() trick . Plz revert .

[segabriel]

I run SyncExampleBenchmarksRunner on my laptop, and before these changes, I received the following results:

i.s.b.e.SyncExampleBenchmarksRunner-timer
             count = 57722532
         mean rate = 959152.09 calls/second
     1-minute rate = 925116.24 calls/second
     5-minute rate = 881552.39 calls/second
    15-minute rate = 870389.79 calls/second
               min = 649.00 nanoseconds
               max = 4780.00 nanoseconds
              mean = 684.39 nanoseconds
            stddev = 158.46 nanoseconds
            median = 660.00 nanoseconds
              75% <= 668.00 nanoseconds
              95% <= 816.00 nanoseconds
              98% <= 899.00 nanoseconds
              99% <= 941.00 nanoseconds
            99.9% <= 3837.00 nanoseconds

And after:

-- Timers ----------------------------------------------------------------------
i.s.b.e.SyncExampleBenchmarksRunner-timer
             count = 122485880
         mean rate = 2035577.46 calls/second
     1-minute rate = 1940461.86 calls/second
     5-minute rate = 1809094.25 calls/second
    15-minute rate = 1776252.26 calls/second
               min = 657.00 nanoseconds
               max = 8453.00 nanoseconds
              mean = 1000.55 nanoseconds
            stddev = 293.65 nanoseconds
            median = 993.00 nanoseconds
              75% <= 1020.00 nanoseconds
              95% <= 1048.00 nanoseconds
              98% <= 1061.00 nanoseconds
              99% <= 1076.00 nanoseconds
            99.9% <= 8453.00 nanoseconds

[artem-v]

I run SyncExampleBenchmarksRunner on my laptop, and before these changes, I received the following results:

Have you noticed one thing: tps rate became larger 2x times, but latency is worsen. That's strange. I run sync exmple runner locally and indeed tps went top 2x times , but latency remained the same, just saying you stats are confusing. But we will merge this PR because it's really improves sync tests. Good job, thanks.

PS on my laptop
Before:

i.s.b.e.SyncExampleBenchmarksRunner-timer
             count = 12287647
         mean rate = 1024352.53 calls/second
     1-minute rate = 1011890.98 calls/second
     5-minute rate = 1009587.47 calls/second
    15-minute rate = 1009188.40 calls/second
               min = 581.00 nanoseconds
               max = 3283.00 nanoseconds
              mean = 635.63 nanoseconds
            stddev = 138.45 nanoseconds
            median = 603.00 nanoseconds
              75% <= 608.00 nanoseconds
              95% <= 833.00 nanoseconds
              98% <= 1080.00 nanoseconds
              99% <= 1261.00 nanoseconds
            99.9% <= 3283.00 nanoseconds

After:

i.s.b.e.SyncExampleBenchmarksRunner-timer
             count = 154712503
         mean rate = 2881602.55 calls/second
     1-minute rate = 2826544.70 calls/second
     5-minute rate = 2764897.07 calls/second
    15-minute rate = 2751162.70 calls/second
               min = 613.00 nanoseconds
               max = 980.00 nanoseconds
              mean = 637.78 nanoseconds
            stddev = 21.83 nanoseconds
            median = 635.00 nanoseconds
              75% <= 639.00 nanoseconds
              95% <= 661.00 nanoseconds
              98% <= 681.00 nanoseconds
              99% <= 693.00 nanoseconds
            99.9% <= 969.00 nanoseconds