What version of FBSD does Yahoo run?

[TM4525 at aol.com]

In a message dated 10/8/04 2:37:38 AM Eastern Daylight Time, 
tedm at toybox.placo.com writes:
Kris and all,

  Sorry for the top post but would you quit feeding the trolls?

Ted Mittelstaedt

PS:  TM, shut up and post some benchmarks proving your side of
the argument.  Not that we would believe them but you deserve to
have to spend some time forging them up.
------------------------------------------------

Ah, so now anyone who questions your data is a Troll. Very convenient.
The entire point of "believability" is the control, and the explanation of
what the test actually tests. Thats the point of having a control, Ted. 
The test that was posted is not "believable" because it doesnt test 
anything  that would actually happen
in the real world. Do you buy a car because it hit 180 on the track? 
Is a car that can hit 190 but gets half the gas milage a better car?

You guys are the ones making the claims that 5.3 is "going to be
so great".  I just wonder how you come to that conclusion if you don't
have any definitive tests. I dont have a release to test, so when its done
I'll test it.

> > - a relatively slow machine (a 1.7Ghz celeron with a 32-bit/33mhz
> > fxp NIC running 4.9) pushes over 250Kpps, so why is your machine, 
> > with seemingly superior hardware, so slow?
> 
> Because traffic is being generated from userland, not from within the
> kernel.
> 
-------------------------------------------------
Actually my traffic generator is in userland too, of course. I guess I'm 
just a better coder than whoever wrote your little benchmark. Or maybe 
the benchmark is too busy calculating stats to do the work its supposed
to be doing. Another variable in the "test".

> For this workload, yes.
> 
> > It also seems that the gap has widened between UP and SMP
> > performance in 5.x. Wasn't one of the goals of 5.x to substantially
> > improve SMP performance?
> 
> Yes, and it's ongoing.  You don't see it on this workload, but there
> are other benchmarks (e.g. mysql select testing) that I don't have to
> hand at the moment, which show the smp benefits of 5.3 more clearly.
> 
> > This seems to show the opposite.
> 
> No, it shows a small increase on SMP and a large increase on UP.
> Anyway, weren't you demanding an email ago that I produce benchmarks
> on UP systems, because no-one really uses SMP?

You must be a democrat Kris, because you always spin what people say
in a way such that is completely wrong when you say it. I said "the 99% of us
who don't use SMP", which is much different from "no one uses SMP", isn't
it? 1% of several million is not "no-one", is it?

Frankly, I didnt expect SMP performance to be so poor in 5.x since improving 
it
is a stated goal. So I guess you recommend that anyone running a network 
server use a single processor? Are the gains in mySQL greater that the 40%
loss in network performance? When mySQL is performaning so aptly, is
the machine capable of handling a network load also?

You (Kris) seem to think I'm asking you these questions, but I'm really not, 
but I guess I'm surprised you keep answering since you clearly don't have 
any of the answers. I'm just hoping someone does, somewhere. Because 
I don't see how you can develop an O/S without benchmarking your specific 
changes along the way.  

The folks at LINUX are guilty of building an O/S to suit their benchmarks. Its
equally disturbing to implement theory without making sure that the theory
works as expected. I just hope that pounding packets through a socket  and 
timing mySQL selects aren't the entirety of your team's arsenal.