incorporated com.codahale.metrics instead of hand-rolled reporting

[leifwalsh]

reduces complexity and provides nice things like latency histograms and windowed average throughput

[kgorman]

Yes please! Tim, please merge it! I am actually in need of CsvReporter.

[kgorman]

Just ran some tests with this version/PR. These stats are perfect. Yes, please merge.

sysbench.updates
             count = 84020
         mean rate = 841.61 calls/second
     1-minute rate = 792.20 calls/second
     5-minute rate = 350.10 calls/second
    15-minute rate = 230.03 calls/second
               min = 0.68 milliseconds
               max = 58.55 milliseconds
              mean = 6.69 milliseconds
            stddev = 3.97 milliseconds
            median = 5.63 milliseconds
              75% <= 7.00 milliseconds
              95% <= 12.76 milliseconds
              98% <= 20.43 milliseconds
              99% <= 25.25 milliseconds
            99.9% <= 40.46 milliseconds

[kgorman]

After testing a bit more with this fork, a couple issues:

• Somehow the ability to observe RUN_TIME_MINUTES is broken, it just runs forever.
• The reporting as listed above should show up at the end only (or an option).
• The sysbench-deletes file systems to always be all zeros(?).

These are just a couple things I see at first glance.

[tmcallaghan]

I'm all caught up on the other PRs, so I'll be looking at how to incorporate this soon.

[kgorman]

Killer, thx Tim!

[leifwalsh]

It should be possible to tighten up the output format by manually getting and printing metrics, I know that was one of your concerns, Tim. 

Cheers,
Leif

On Thu, Feb 12, 2015 at 12:34 PM, Kenny Gorman [email protected]
wrote:

Killer, thx Tim!

Reply to this email directly or view it on GitHub:
#11 (comment)