Calculate throughput based on average value

Add coordinator patch

BUG=

[Jiangfeng, 2017-12-07 19:28:04]

I found wrong logic and see the comments that I write to the twos lines in the
code.

https://codereview.appspot.com/333230043/diff/1/src/java/org/apache/cassandra...
File src/java/org/apache/cassandra/service/AbstractReadExecutor.java (right):

https://codereview.appspot.com/333230043/diff/1/src/java/org/apache/cassandra...
src/java/org/apache/cassandra/service/AbstractReadExecutor.java:213:
SpeculativeRetryParam retry = cfs.metadata().params.speculativeRetry;
I suggest you save the current percentile parameter of speculative retry as a
static variable. And write like "SpeculativeRetryParam retry =
SpeculativeRetryParam.percentile(thisvariable);"

And then change the static variable at the end of ML calculation.

https://codereview.appspot.com/333230043/diff/1/src/java/org/apache/cassandra...
src/java/org/apache/cassandra/service/AbstractReadExecutor.java:236: retry =
SpeculativeRetryParam.percentile(Float.parseFloat(output));
If retry is set here, then, the readexecutor of the only request is set with our
calculated speculative retry parameter. All other read request is still set by
Line 213 "SpeculativeRetryParam retry =
cfs.metadata().params.speculativeRetry;".