Re: inconsistent implementation of 'cumulative distribution function' for Exponential Distribution

[Jonathan Ellis <jb...@gmail.com>]

I don't know which is correct.  It's possible that both are, because
of differences in how they are used.

Created https://issues.apache.org/jira/browse/CASSANDRA-2597 to look into this.

On Sat, Apr 2, 2011 at 7:12 AM, pprun <pz...@gmail.com> wrote:
> Hi Cassandra DEVs,
>
> I just found there are two implementations of 'cumulative distribution
> function' for Exponential Distribution and there are inconsistent :
>
> *FailureDetector*
> org.apache.cassandra.gms.ArrivalWindow.p(double)
>    double p(double t)
>    {
>        double mean = mean();
>        double exponent = (-1)*(t)/mean;
>        return *Math.pow(Math.E, exponent)*;
>    }
>
> *DynamicEndpointSnitch*
> org.apache.cassandra.locator.AdaptiveLatencyTracker.p(double)
>    double p(double t)
>    {
>        double mean = mean();
>        double exponent = (-1) * (t) / mean;
>        return *1 - Math.pow( Math.E, exponent);*
>    }
>
> According to the  Exponential Distribution cumulative distribution function
> definition<http://en.wikipedia.org/wiki/Exponential_distribution#Cumulative_distribution_function>,
> the later one is clear, however,
> Could you please explain a bit for the implementation of FailureDetector?
>
> Thanks*,
> *pzgyuanf*
> *
>



-- 
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of DataStax, the source for professional Cassandra support
http://www.datastax.com