Re: odd -CURRENT performance issue

From: John Baldwin <jhb_at_FreeBSD.org>
Date: Tue, 8 Feb 2005 13:26:14 -0500
On Monday 07 February 2005 06:49 pm, Don Lewis wrote:
> 	time make all-depends-list
> ten times in a row in /usr/ports/x11/gnome2.  The system was freshly
> booted, and other than a niced setiathome, the system was idle.  I got
> the following results:
>
>       122.63 real        29.83 user        66.29 sys
>       117.86 real        29.58 user        66.55 sys
>       119.89 real        29.80 user        67.27 sys
>       121.68 real        30.31 user        67.55 sys
>       122.10 real        30.41 user        67.66 sys
>       120.90 real        29.67 user        68.24 sys
>       121.21 real        30.27 user        67.69 sys
>       219.85 real        30.35 user       148.12 sys
>       312.54 real        30.49 user       224.79 sys
>       312.83 real        30.23 user       225.83 sys
>
> The last three lines are rather strange ...
>
> Kernel and world are 6.0-CURRENT, built from February 3rd sources.
> WITNESS, INVARIANTS, DEBUG_VFS_LOCKS, and SMP enabled, and it is fairly
> close to GENERIC.

WITNESS has especially poor scalability issues and can get much worse as more 
lock classes and relationships are added to its tree due to how it tries to 
rebalance the tree every time it adds a new relationship (and each rebalance 
is O(n) I think).  If you turn off witness via sysctl debug.witness.watch=0,. 
does the same weirdness persist?

-- 
John Baldwin <jhb_at_FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org
Received on Tue Feb 08 2005 - 17:32:20 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:27 UTC