Availability Reporting Inaccuracy

Question

We have been working on creating some availability reports for our remote sites to ensure SLA’s are being met. However we noticed that the built in reporting within Orion was not accurate and I wanted to share this with others in case you depend upon these reports. We confirmed the inaccuracies by going to the Node page and looking at the Availability Statistics and comparing it to the Year to Date report within Orion and found the numbers did not match up. In my example, I have a node that has under the Availability Statistics an up time for this year of 99.502% however the Orion report shows 99.99%.

So the question is why the Orion information isn’t matching up, and it comes from how Orion stores the information. All availability information in Orion is placed within three tables which are ResponseTime.Detail, ResponseTime.Hourly, and ResponseTime.Daily. The detail table consists of every response time request for the last seven days, if you have the polling set to the default of 120 seconds, you will have ~5000 entries in this table for your one node. After seven days, Orion does some clean up and averages these statistics out into an hourly average for the last 30 days which is around ~500 entries in the table. Finally after thirty days, Orion does some cleanup into a daily average for the last 365 days before purging the data. If you are doing a year to date report and today is March 18th, you are only going to have ~40 entries.

Orion now takes the ~5000 + ~500 + ~40 entries and averages them all out with the same weighted value for each entry, the total being ~5540.  The ~5000 detailed entries for the last seven days are going to grossly out-weigh your ~40 daily entries and create a false availability percentage.

There are ways to correct this on your own but you will most likely want to have a DBA help you with this. We are having our DBA create a new table with a nightly job that will take the previous day’s detailed availability, average them out into a daily average. We are then going to build a report off of that table to provide us with our Availability report.

Let me know if you have any questions.

Thanks

clubjuggle · Answer

I have noticed the same thing as well, and brought it to support's attention about a year ago. Bottom line is the canned availability reports do NOT function as advertised, and should not be relied upon to accurately report the statistics they claim to convey. Hopefully this thread will encourage Solar Winds to make correction of this issue a priority.

bsciencefiction.tv · Answer

We have noticed the same thing with an availability report set  up for Last Month versus specifying the dates of last month.  When you select show SQL they have the same exact query, but both reports have different data.