DragonFly vs FreeBSD scheduler

Alie Tan

unread,

Nov 3, 2012, 9:18:55 AM11/3/12

to freebsd...@freebsd.org

Hi,

No offence, just curious about scheduler and its functionality.

What is the different between this two that makes FreeBSD performance far
behind DragonFly BSD? http://www.dragonflybsd.org/release32/

Regards,
Alie T
_______________________________________________
freebsd...@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-curre...@freebsd.org"

Mark Felder

unread,

Nov 3, 2012, 10:17:52 AM11/3/12

to Alie Tan, freebsd...@freebsd.org

On Sat, 3 Nov 2012 21:18:55 +0800
Alie Tan <al...@affle.com> wrote:

> Hi,
>
> No offence, just curious about scheduler and its functionality.
>
> What is the different between this two that makes FreeBSD performance far
> behind DragonFly BSD? http://www.dragonflybsd.org/release32/
>

I don't have any details but I do know that Dragonfly has been putting a lot of work into their scheduler. Hopefully some of that will trickle back our way.

Alfred Perlstein

unread,

Nov 3, 2012, 11:01:51 AM11/3/12

to Alie Tan, freebsd...@freebsd.org

On 11/3/12 6:18 AM, Alie Tan wrote:
> Hi,
>
> No offence, just curious about scheduler and its functionality.
>
> What is the different between this two that makes FreeBSD performance far
> behind DragonFly BSD? http://www.dragonflybsd.org/release32/

Looks like a few specific benchmarks that DragonFly aimed to do well at
that we were unawares of.

Not sure, didn't see DragonFly sharing the results with us until this
paper was published although I may have missed that.

-Alfred

Mehmet Erol Sanliturk

unread,

Nov 3, 2012, 11:19:46 AM11/3/12

to Alfred Perlstein, freebsd...@freebsd.org, Alie Tan

On Sat, Nov 3, 2012 at 8:01 AM, Alfred Perlstein <bri...@mu.org> wrote:

> On 11/3/12 6:18 AM, Alie Tan wrote:
>
>> Hi,
>>
>> No offence, just curious about scheduler and its functionality.
>>
>> What is the different between this two that makes FreeBSD performance far

>> behind DragonFly BSD? http://www.dragonflybsd.org/**release32/<http://www.dragonflybsd.org/release32/>

>>
>
> Looks like a few specific benchmarks that DragonFly aimed to do well at
> that we were unawares of.
>
> Not sure, didn't see DragonFly sharing the results with us until this
> paper was published although I may have missed that.
>
> -Alfred
>
>
>

O. Hartmann

unread,

Nov 3, 2012, 11:29:33 AM11/3/12

to freebsd...@freebsd.org

Am 11/03/12 16:01, schrieb Alfred Perlstein:

> On 11/3/12 6:18 AM, Alie Tan wrote:
>> Hi,
>>
>> No offence, just curious about scheduler and its functionality.
>>
>> What is the different between this two that makes FreeBSD performance far
>> behind DragonFly BSD? http://www.dragonflybsd.org/release32/
>
> Looks like a few specific benchmarks that DragonFly aimed to do well at
> that we were unawares of.

What is "specific" supposed to mean? It doesn't seem to be very
"specific". This benchmark reflects a general tendency which can also be
observed by the vast of benchmarks "Phoronix" performed.

I guess FreeBSD has been benchmarked with ULE. ULE does have issues and
it's obvious, that ULE performs in specific situations better than the
legacy and old (but after so many years still competetive) BSD scheduler.

signature.asc

O. Hartmann

unread,

Nov 3, 2012, 11:36:59 AM11/3/12

to freebsd...@freebsd.org

Am 11/03/12 15:17, schrieb Mark Felder:

> On Sat, 3 Nov 2012 21:18:55 +0800
> Alie Tan <al...@affle.com> wrote:
>
>> Hi,
>>
>> No offence, just curious about scheduler and its functionality.
>>
>> What is the different between this two that makes FreeBSD performance far
>> behind DragonFly BSD? http://www.dragonflybsd.org/release32/
>>
>
> I don't have any details but I do know that Dragonfly has been putting a lot of work into their scheduler. Hopefully some of that will trickle back our way.

Obviously they made the right decissions, but a single benchmark with a
DB server like postgresql doesn't tell the whole story. Let's see what
Phoronix will come up with. I'd like to see some more benchmarks of
DragonFly 3.2.

I doubt that the DragonFly scheduler approaches will go/flow easily into
FreeBSD. But I'd like to see it, even dumping ULE for a better approach.

signature.asc

Jeff Roberson

unread,

Nov 3, 2012, 2:44:55 PM11/3/12

to O. Hartmann, freebsd...@freebsd.org

It's not the scheduler. It's lock contention in the vm and buffer cache.
The scheduler can only schedule what is runnable. We are working to
address this problem.

Thanks,
Jeff

Alfred Perlstein

unread,

Nov 3, 2012, 11:30:30 PM11/3/12

to Samuel J. Greear, freebsd...@freebsd.org, Alie Tan

On 11/3/12 7:02 PM, Samuel J. Greear wrote:
>> Looks like a few specific benchmarks that DragonFly aimed to do well at that
>> we were unawares of.
>>

> Unawares of? http://bsd.slashdot.org/story/08/03/06/1313218/freebsd-70-bests-linux-in-smp-performance
> The FreeBSD project made hay with sysbench and pgbench not that long ago.

>
>> Not sure, didn't see DragonFly sharing the results with us until this paper
>> was published although I may have missed that.
>>

> Is the DragonFly project under some obligation to share its results
> with the FreeBSD developers?

My takeaway is that this conversation is dumb.

Samuel J. Greear

unread,

Nov 3, 2012, 10:02:01 PM11/3/12

to Alfred Perlstein, freebsd...@freebsd.org, Alie Tan

> Looks like a few specific benchmarks that DragonFly aimed to do well at that
> we were unawares of.
>

Unawares of? http://bsd.slashdot.org/story/08/03/06/1313218/freebsd-70-bests-linux-in-smp-performance
The FreeBSD project made hay with sysbench and pgbench not that long ago.

> Not sure, didn't see DragonFly sharing the results with us until this paper
> was published although I may have missed that.
>

Is the DragonFly project under some obligation to share its results
with the FreeBSD developers?

Reply all

Reply to author

Forward