[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Data precision

From: Martin Pala
Subject: Re: Data precision
Date: Fri, 16 May 2014 08:57:42 +0200

The precision and data retention is hardcoded and automatic. It is easy to 
change the retention and keep the data for longer time, but it will be 
suboptimal for chart rendering ... for example 1 month with 1 minute precision 
would mean 44640 datapoints (single service) vs. current ~ 62 datapoints - if 
let's say 10 services will be rendered to the same chart, it will be ca. 0.5M 
datapoints which won't be very fast on desktop, not to mention phones or 
tablets (even though it is could be possible to switch to lower resolution for 
mobile devices and keep high resolution on desktop only). It will be better to 
add separate "maximum" based dataset to current "average" dataset.


On 15 May 2014, at 23:53, Francisco Reyes <address@hidden> wrote:

> Is there any way to have a different precision from the current defaults?
> The existing setup
> Last Seconds Snapshot
> Last Hour 1 Minute
> Last Day 15 Minutes
> Last Week 2 Hours
> Last Month 12 Hours
> Last Six Months 1 Day
> Last Year 7 Days
> Last Three Years 14 Days
> Is not effective for instances were there is mostly a very steady stream of 
> usage and then there are spikes.
> The averages completely and totally hide what would be extremely valuable 
> information.
> Today I had a system crash.
> I got to the machine within the hour.
> Apache had a peak of 90+ instances and it was steady at 80+ until the machine 
> crashed.
> Looking at the hour stats showed the issue..
> Now several hours later looking at the day stat it is showing the peak as 
> 63.. If I switch to week it shows 20.  We haven't had mmonit for a month, but 
> I am sure it will be near useless in the month graph.
> The existing algorithm and retention policies may save space, but make mmonit 
> to be significantly less effective for cases of constant load with sporadic 
> spikes. The averaging completely hides the spikes.
> I would far rather throw the data in Postgresql.. or have  just a few months 
> of retention... than to have the current scenario which makes stats, for this 
> particular scenario, useless after a single day.
> --
> To unsubscribe:

reply via email to

[Prev in Thread] Current Thread [Next in Thread]