Subject: Re: FreeBSD Bus DMA
To: None <dyson@freebsd.org>
From: Perry E. Metzger <perry@piermont.com>
List: tech-kern
Date: 06/12/1998 00:00:22
"John S. Dyson" writes:
> Perry E. Metzger said:
> > "John S. Dyson" writes:
> > > Funny, that is why our build times are 3X faster on FreeBSD, for NetBSD
> > > based products than on NetBSD, right?  We even have a development
> > > team avoiding *BSD all together because of NetBSD/X86 performance
> > > problems.
> > 
> > You are claiming that FreeBSD is three times faster than NetBSD on
> > identical hardware? THREE TIMES?
> > 
> > My last check determined that, so far as I could tell, NetBSD and
> > FreeBSD were more or less within a few percent of each other in
> > performance. NetBSD is faster at some things, slower at others --
> > certainly no real measurable difference.
>
> How are you measuring it?  Probably with unloaded system or LL benchmarks.

No. I was doing real work. Doing things like trying to compile TeX and 
such. I couldn't find any significant difference -- maybe a couple
percent at best, and not always in your favor, either.

You are correct that I didn't use machines pushed past the edge of
normal tolerable load.

If you are talking about a machine being forced to page for its life,
I wouldn't call that a reasonable system to be doing work on, under
either FreeBSD or NetBSD, and most other users wouldn't either. None
the less, I suspect in -current we are very respectable in this
regard.

> > If you feel there is a difference, though, feel free to present a
> > benchmark.
>
> Just do a massive compile, with FreeBSD set up correctly...

Fine. Get me the numbers. Build identically configured Gnu Emacs on a
NetBSD box and on the same box running FreeBSD. Configure FreeBSD any
way you like -- I don't care much -- just make sure NetBSD isn't being 
configured to do something unfair, like page over a PPP link or
something.

> You can do it.  It is pretty obvious if you actually try it.  Our
> people are happy about the change in compile times (it is massive.)

It is, in fact, nonexistant from what I can tell.

> > "Huh?"
> > 
> > Please, feel free to show us the numbers.
>
> Experience by others trying to get software built.  In fact, it
> was NetBSD :-).

"Huh"???

Give us the numbers. Numbers beat words.

> > 
> > Might I ask, btw, why the "FreeBSD" Alpha folks are using a NetBSD
> > kernel?
>
> Not for long.   Purely a bootstrapping issue.  Did the NetBSD people
> ever use SunOS or Solaris?

No, we never booted or distributed SunOS or Solaris kernels claiming
they were NetBSD kernels, and never used SunOS or Solaris kernels with
NetBSD Userlands saying this was "NetBSD/Sparc". We never put SunOS or
Solaris kernels into our source tree as "NetBSD/Sparc".

It is true that we have used other operating systems during bootstrap, 
but we haven't called them our own, no.

> > > FreeBSD isn't perfect, but also claiming somehow that only you have
> > > magical and more correct solutions seems to show that working with
> > > you isn't going to happen.
> > 
> > I'm not claiming we have a monopoly on brains. I'm claiming FreeBSD
> > has never been ported to anything other architecture from its Wintel
> > base. Do you dispute this?
> >
> I am going to frame your first sentence, simply because it is
> seldom heard from a NetBSD person.  A FreeBSD kernel is soon to run
> on an Alpha,

"Soon"?

When a kernel that isn't a NetBSD kernel is running, and your port is
finished enough for users (that is, its something people would care to 
use day to day), then you will have finished doing the work. Right
now, you've yet to boot a FreeBSD/Alpha kernel that you actually wrote.

> but why should you care, since FreeBSD is so inferior, right?

As it doesn't exist, how can one comment on its merits?

> > > We certainly don't want to be constrained
> > > by the limitations of an OS that has lots of ports...  sort of...
> > 
> > Sort of?
>
> Well I know of and used a couple of NetBSD ports that do not
> meet the standards that I feel would be sufficient for FreeBSD...
> One of them is the X86 port,

So you are claiming we have no functioning ports?

I don't know what your "standards" are. I currently have a firewall in
production at a Big New York Company that's been running for over a
year without rebooting or having any observable performance issues. It
is running NetBSD/i386 on what would currently be considered rinky
dink hardware, but it is functioning just fine at managing a large
amount of traffic. It's doing fine. I'm sure FreeBSD would have been no
worse -- but you would be utterly disingenuous to claim it would have
somehow done better.

> and the other one, I will not say,
> because I like and respect the ports-meister so much.

That is what we call a smear in the buisness. "I know of fifteen
communists who work for FreeBSD, but I won't name them because I don't 
want to cause trouble".

Say what you mean.

> I have been helping with a little NetBSD hacking, and frankly, I would
> be very embarassed about the way the code works.  Of course, we fixed
> most of those problems in FreeBSD.

Embarassed? By what, exactly? What in our code base is "embarassing"?

Or is this like your benchmarks?

Perry