What version of FBSD does Yahoo run?

[Kris Kennaway]

On Thu, Oct 07, 2004 at 07:04:10AM -0700, Kris Kennaway wrote:
> On Thu, Oct 07, 2004 at 09:55:37AM -0400, TM4525 at aol.com wrote:
> > In a message dated 10/6/04 6:47:04 PM Eastern Daylight Time, 
> > kris at obsecurity.org writes:
> > On Wed, Oct 06, 2004 at 04:42:24PM -0400, Bigelow, Andrea L. wrote:
> > > Where's the documentation? I'd like to see this for myself.  
> > 
> > There is none, because Mr./Ms. TM4525 is making up his/her "facts" to
> > suit their assertion.  The last time this claim was made it was
> > refuted and TM4525 promised to go away and check 5.3 performance.
> > 
> > Kris
> > ----------------------------------
> > 
> > Actually, Kris, it wasn't "refuted", you said that the exceptionally poor 
> > performance was "expected" until 5.3 was released, and implied that 
> > anyone who expected good performance was making a fool of themselves.
> > 
> > Search google groups for "freebsd 5.2 performance woes" and sort by date 
> > to see my test details and subsequent comments by Kris and the other 
> > FreeBSD Spin Doctors.
> > 
> > My tests are very controlled, and my "assertion" is a result of exceptionally
> > poor performance in the test. And no-one "refuted" my results. More 
> > like jockeying to save face.
> > 
> > Nor did I "promise to go away". I promised to test 5.3 and post 
> > the results.
> 
> We're waiting..5.3 is in beta and ready for your tests.  Other
> benchmarks show very good results compared to 4.x.

Here's one benchmark, showing UDP packet/second generation rate from
userland on a dual xeon machine under various target loads:

Desired Optimal 5.x-UP  5.x-SMP 4.x-UP  4.x-SMP
50000   50000   50000   50000   50000   50000
75000   75000   75001   75001   75001   75001
100000  100000  100000  100000  100000  100000
125000  125000  125000  125000  125000  125000
150000  150000  150015  150014  150015  150015
175000  175000  175008  175008  175008  169097
200000  200000  200000  179621  181445  169451
225000  225000  225022  179729  181367  169831
250000  250000  242742  179979  181138  169212
275000  275000  242102  180171  181134  169283
300000  300000  242213  179157  181098  169355

i.e. it shows a 33% improvement on UP machines, and 6% on SMP between
4.x and 5.3.  (Of course, kernel packet generation is much faster than
userland, but that's not what is benchmarked here.)

SMP in 5.3 does a lot better in benchmarks of other types of
workloads, for example mysql with the "supersmack" stress tool.  I
don't have those numbers to hand right now though.

Of course, there are lots of other things you could try to benchmark,
and there is certainly a lot of optimization work remaining to be
done.  The first step in optimizing is to find a good test case that
clearly demonstrates a problem, and run it under controlled
conditions.  But this shows that 5.3 is clearly a good start along
that path, and is a significant improvement over 4.x and older 5.x
releases.  You should expect further performance improvements in the
5.x branch over the coming months, as the focus of development shifts
from infrastructure to optimization.

Kris
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-questions/attachments/20041007/a0c50a22/attachment.bin