On 01/17/2013 08:19 PM, Rahul Sundaram wrote:
On 01/17/2013 07:00 PM, M. Edward (Ed) Borasky wrote:
Yes, I'd veto btrfs as the default as well. I lost a huge chunk of
data on a btrfs partition a while back, with *no* diagnostics,
..
You don't have the power to veto it although highlighting critical bugs
will be beneficial. If you are going to highlight them, do post
bugzilla links rather than descriptions of the issues.
I used btrfs on my personal desktop for about a year now and it works
fine for me, other than a nagging suspicion that there's an occasional
performance problem at high load. I can't say anything intelligent about
it because I coudn't figure out how to measure what I feel, but the
system seems sluggish during high-impact events like 'yum update' and/or
firefox with gazillion tabs open. The simple diagnostics (top, ps,
iotop) don't seem to point to anything definitive: it essentially shows
reasonable CPU load, relatively low IO rate and load average in single
digits, with occasional significant (second-like) latency for simple
things like window exposes and terminal keystrokes.
I tried to come up with ways of measuring and quantifying, e.g. with
systemtap, but it didn't get me anywhere. It would be great if someone
familiar with performance measurements could write up how to narrow down
issues like that in a complex desktop situation where X interacts with
IO, networking, virtual memory, etc.
Even better, maybe there's a way to have a kernel-side or Xserver-side
component that automatically detects user interaction latency and logs
warnings that would help in pinpointing the causes. I assume that
latency can be detected by comparing low-level event times (USB traffic
from the mouse, etc), to high-level event times like window redraws.
--
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/devel
