Paolo Valente
unread,Jun 15, 2013, 10:37:15 AM6/15/13Sign in to reply to author
Sign in to forward
You do not have permission to delete messages in this group
Sign in to report message
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to bfq-i...@googlegroups.com
Hi,
also this new revision is ready, at last. It is a set of fixes, the most important of which preserves fairness also for synchronous queues. We got to this fix thanks to a fairness problem reported by Andrea Parri.
Interestingly, this fix, together with a fix related to EQM, further improves the stability of the start-up times of short-to-middle-size applications. There are still the usual latency-fluctuations problems with larger applications, such as LibreOffice Writer.
We will go on trying to better understand the cause of those problems as well.
As usual, Arianna prepared a ton of patches and patch-sets for many kernel versions.
If you are interested into further details, here is the changelog:
v6r2:
- Fairness fix: the case of queue expiration for budget timeout is
now correctly handled also for sync queues, thus allowing also
the processes corresponding to these queues to be guaranteed their
reserved share of the disk throughput.
- Fixed a bug that prevented group weights from being correctly
set via the sysfs interface.
- Fixed a bug that cleared a previously-set group weight if the
same value was re-inserted via the sysfs interface.
- Fixed an EQM bug that allowed a newly-started process to skip
its initial weight-raising period if its queue was merged before
its first request was inserted.
- Fixed a bug that preserved already-started weight-raising periods
even if the low_latency tunable was disabled.
- The raising_max_time tunable now shows, more user-friendly, the
maximum raising time in milliseconds.
Any feedback is welcome,
Paolo