User: Password:
|
|
Subscribe / Log in / New account

No-I/O dirty throttling

No-I/O dirty throttling

Posted Sep 8, 2011 8:41 UTC (Thu) by kevinm (guest, #69913)
Parent article: No-I/O dirty throttling

From an application developer / system administrator perspective, it would be nice if processes that are sleeping due to being throttled would appear in a state other than 'S', which implies voluntary blocking. Perhaps 'E' for "Enforced Sleep" (even just 'D' state would be good enough, if a new process state would be deemed to be a breaking ABI change).


(Log in to post comments)

No-I/O dirty throttling

Posted Sep 8, 2011 13:19 UTC (Thu) by corbet (editor, #1) [Link]

The current patches just put the process into a normal interruptible sleep, so it won't appear any different. Remember, though, that the maximum sleep period is measured in milliseconds, so it's not something you would expect to see all that often in a ps display.

No-I/O dirty throttling

Posted Nov 9, 2011 4:21 UTC (Wed) by hamjudo (guest, #363) [Link]

If something has a 20% duty cycle, and you are randomly sampling, you expect to see the something in that special state in about 1 out of 5 samples. It doesn't matter if the average cycle time is 37 milliseconds or 20 microseconds.

If the throttling code, only throttles back 20%, then things are probably running just fine. A nightmare situation, such as running a web browser from 2011 swapping to a thumbdrive designed in 2006, would more likely have the process generating dirty pages for less than 10ms, then being put to sleep for 200ms. This would give the observer less than a 5% chance of witnessing the process while it was "awake".


Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds