[haiku-bugs] Re: [Haiku] #10454: New scheduler: substantial performance drop

  • From: "stippi" <trac@xxxxxxxxxxxx>
  • Date: Wed, 05 Feb 2014 14:42:38 -0000

#10454: New scheduler: substantial performance drop
-----------------------------+----------------------------
   Reporter:  stippi         |      Owner:  pdziepak
       Type:  task           |     Status:  in-progress
   Priority:  high           |  Milestone:  R1
  Component:  System/Kernel  |    Version:  R1/Development
 Resolution:                 |   Keywords:
 Blocked By:  10487          |   Blocking:
Has a Patch:  0              |   Platform:  All
-----------------------------+----------------------------

Comment (by stippi):

 It is hard to tell whether the short freezes are related. I have
 definitely seen freezes in mouse events (1 - 3 secs) on this machine from
 time to time (before the scheduler merge). These are USB related, the
 syslog has entries similar to "USB port reset". It gets really bad when I
 connect my Wacom Intuos 2 tablet. This is definitely unrelated. The short
 delays while typing only happen very infrequently and I can say that I got
 aware of them after the scheduler merge, but they might have been there
 before and might be related to the mouse freezes. I can keep a tail -F
 syslog open the next time I do some Haiku work. That should clear up at
 least this issue.

 My musings about the mysterious latency might not make much sense. I told
 you that when I run the WonderBrush prototype which uses as much render
 threads as there are CPU cores, I see those cores all pegged. They
 wouldn't be if threads are not always allowed to run when they can.

 To think about the problem as something thread migration related still has
 some question marks for me. A new task should be scheduled immediately,
 even when it is short lived. It is either scheduled on a CPU with less
 load, or on a CPU which is already full, in which case it runs later than
 it could. If this theory is based on facts, then there should always be at
 least one CPU under full load. You are saying this may still be the case,
 only that Activity Monitor misleads by measuring load over such a long
 interval?

--
Ticket URL: <https://dev.haiku-os.org/ticket/10454#comment:11>
Haiku <https://dev.haiku-os.org>
Haiku - the operating system.

Other related posts: