Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Accumulo, mail # user - Computations behind JMX Metrics


Copy link to this message
-
Re: Computations behind JMX Metrics
Eric Newton 2013-03-01, 14:35
Yes, the average metrics obtained via JMX are exponentially smoothed.

You can get more accurate counters via a thrift call to the master, and you
can trace specific calls to get fine-grained time information for all parts
of a specific transaction.

-Eric
On Fri, Mar 1, 2013 at 9:26 AM, Brandon Wilson <[EMAIL PROTECTED]
> wrote:

> I have a question regarding the JMX metrics exposed by Accumulo. I believe
> the metrics that are reported as averages (e.g., ScanAvgTime) are computed
> as weighted averages over all time, as opposed to a arithmetic mean of a
> set of scan times over a given time window.  Am I correct that the weighted
> average calculation is actual the weighted average since the tserver
> started?
>
> The average formula I suspect is being use is here:
> http://grepcode.com/file/repo1.maven.org/maven2/org.apache.accumulo/accumulo-server/1.4.0/org/apache/accumulo/server/metrics/AbstractMetricsImpl.java#AbstractMetricsImpl.Metric.addAvg%28long%29<https://myremote.mitre.org/owa/,DanaInfo=imc.mitre.org,SSL+redir.aspx?C=c6-KKQhSR0qXLqfNtF0WFWOUgfkZ6s8ILg3UmiJgkZ-j_nin7It-AddWEsRTPiKlFa7QDfn4H8s.&URL=http%3a%2f%2fgrepcode.com%2ffile%2frepo1.maven.org%2fmaven2%2forg.apache.accumulo%2faccumulo-server%2f1.4.0%2forg%2fapache%2faccumulo%2fserver%2fmetrics%2fAbstractMetricsImpl.java%23AbstractMetricsImpl.Metric.addAvg%2528long%2529>
>
> Am I on the right track here?
>
> -Brandon
>