[Kea-users] Results of Kea performance survey posted

[Victoria Risk]

As promised, here are the results of the survey we did last week on Kea performance.  

I kept the data tables in Survey Monkey, to avoid posting .pdfs to this list.  Go to this URL to see a summary of the results. https://www.surveymonkey.com/results/SM-RWNVL7SFV/  Click on any blue links to see individual responses or comments. 

The responses were bi-modal; respondants had either less than 5 thousand end points, or 50 - 100,000 end points.  It turns out that the most interesting points are only seen by comparing the smaller and larger deployments, which unfortunately the canned Survey Monkey report above doesn’t do.

For the smaller deployments (<5K endpoints) observed throughput was <50 LPS. That is not unexpected, because it would be hard to get higher lease rates without more clients. For the >10K endpoints deployments, throughput ranged from 50 - 500 LPS.

The most popular lease durations overall were 1-12 hours, followed by 12 - 24 hours. However, for the larger deployments, the most popular choice was 15 minutes to 1 hour. 

Twelve of the 20 respondants are using MySQL, and six are using memfile for leases. 

Eight of the 20 have created their own HA model, using a shared lease backend, instead of using the open source HA hook. 5 of 8 large deployments are using this configuration, which today is not optimal for performance. This indicates to me that we should prioritize that scenario when we tackle performance in Kea 1.7.

Four respondants are using their own non-ISC developed hooks.  That is the same number who are running Kea 1.2 (long since EOL) - coincidence?? 

Two to Four cores is the most common CPU configuration, with half of the respondents running Kea in a VM.  
Overall, 3 of the 20 respondants were dissatisfied with Kea performance, two of those were larger deployments.

This was a pretty small survey and it won’t be featuring in any conference presentations, but the feedback is helpful to us as we plan for improving Kea performance.  Thank you to everyone who took the time to complete the survey.

Vicky

---------------
Victoria Risk
Product Manager
Internet Systems Consortium
vicky at isc.org





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/kea-users/attachments/20190408/ecc59259/attachment.htm>