Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 2 Jul 2007 12:02:41 -0700 (PDT)
From:      Jeff Roberson <jroberson@chesapeake.net>
To:        Thomas Sparrevohn <Thomas.Sparrevohn@btinternet.com>
Cc:        freebsd-current@freebsd.org
Subject:   Re: New SCHED_SMP diff.
Message-ID:  <20070702120147.F552@10.0.0.1>
In-Reply-To: <200707021956.01128.Thomas.Sparrevohn@btinternet.com>
References:  <20070701224452.I552@10.0.0.1> <499c70c0707020950p180d88e1q109b1a1c820d8283@mail.gmail.com> <200707021956.01128.Thomas.Sparrevohn@btinternet.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 2 Jul 2007, Thomas Sparrevohn wrote:

> On Monday 02 July 2007 17:50:44 Abdullah Ibn Hamad Al-Marri wrote:
>> On 7/2/07, Jeff Roberson <jroberson@chesapeake.net> wrote:
>>> I fixed a major bug in SCHED_SMP that impacted some users causing bad
>>> performance and invalid load counts.  Attilio also added support for i386
>>> based machines.  I have tested on UP which works although INVARIANTS and
>>> WITNESS don't work properly on UP kernels.  Single processor SMP kernels
>>> work fine.
>>>
>>> This diff also includes another patch that I intend to commit to current
>>> which significantly improved write performance for mysql by removing
>>> all use of Giant in fcntl().
>>>
>>> Please report any stability or performance problems with this patch.  The
>>> more positive reports I get the more likely it will go into 7.0.
>>>
>>> Thanks,
>>> Jeff
>>
>> Jeff,
>>
>> This makes my Intel C2D E6660 get really busy, and MySQL 5.0.41 from
>> ports run faster :)
>>
>> Thanks for making such great SCHED, when will you make it in the tree? :)
>>
>
> Hmm - Weird it gives negative results on my C2D XQ6700 - I ran four
> tests using "make world" benchmark - I am using ZFS for all filesystems
> except root - however I would not think that matters - I tested with "-j 8"
> and "-j 4" but it did not change the overall result
>

Hi, can you try sysctl kern.sched.steal_busy=1?  Please tell me if that 
improves things.  This will resolve short-term load imbalances by stealing 
threads from other processors.

Thanks,
Jeff

>
> SCHED_SMP - rm -rf /usr/obj/* && /usr/bin/time -l make -j 4 buildworld
>
> 1272.53 real      2482.83 user       945.08 sys
> 1268.45 real      2483.46 user       950.95 sys
> 1274.19 real      2481.91 user       953.19 sys
> 1292.27 real      2482.59 user       957.37 sys
>
>   2423814  voluntary context switches
>    885934  involuntary context switches
>
>
> SCHED_4BSD - rm -rf /usr/obj/* && /usr/bin/time -l make -j 4 buildworld
>
> 1125.28 real      2675.17 user      1055.18 sys
> 1124.74 real      2684.95 user      1057.64 sys
> 1123.76 real      2678.64 user      1061.89 sys
> 1120.68 real      2676.78 user      1061.03 sys
>
>
>   2587885  voluntary context switches
>    708274  involuntary context switches
>
>
>
>
> _______________________________________________
> freebsd-current@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe@freebsd.org"
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070702120147.F552>