You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Roman Puchkovskiy (Jira)" <ji...@apache.org> on 2022/04/21 07:29:00 UTC

[jira] [Created] (IGNITE-16878) Use exponential moving average to estimate checkpoint write speed

Roman Puchkovskiy created IGNITE-16878:
------------------------------------------

             Summary: Use exponential moving average to estimate checkpoint write speed
                 Key: IGNITE-16878
                 URL: https://issues.apache.org/jira/browse/IGNITE-16878
             Project: Ignite
          Issue Type: Improvement
          Components: persistence
            Reporter: Roman Puchkovskiy


The speed-based throttling (implemented inĀ  SpeedBasedMemoryConsumptionThrottlingStrategy) uses an estimation of checkpoint write speed (in pages per second) to decide whether to throttle (and how hard) or not.

Currently, this estimation is obtained by taking a moving average for the current checkpoint and up to 3 preceeding checkpoints. This should work well if the checkpoint writing speed is stable. But if this speed changes a lot (probably, due to some other disk activity, like rotating the WAL or activity of other applications), this change only becomes noticed after a substantial amount of time.

The idea is to use exponential moving average https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average to make the metric note the changes faster. Other weightings can also be considered.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)