7520 Chipset support in 4.x

[JohnsoBS at vicksburg.navy.mil]

> -----Original Message-----
> From: TM4525 at aol.com [mailto:TM4525 at aol.com]
> Sent: Saturday, October 30, 2004 9:13 PM
> To: questions at freebsd.org
> Subject: Re: 7520 Chipset support in 4.x
> 
> 
> > > Many of the new MBs from such tiny vendors as Dell and Supermicro
> > > are based on the 7520, and word is that FreeBSD 4.x 
> doesn't support
> > > it. Is support forthcoming?
> > 
> > We have 2 Dell PowerEdge 1850 servers which have the e7520 
> chipset. They
> > hang consistently in 4.10-RELEASE and below whenever there 
> is high network
> > or disk utilization. We have not been able to get any 
> debugging info.
> > After upgrading to 4.10-STABLE a couple of weeks ago, they 
> no longer hang,
> > but they are _really_ slow to perform network and disk operations.
> > 
> > They work fine in FreeBSD 5.3, but unfortunately our 
> applications do not
> > run without recompiling. We do not want to change our environment to
> > support different binaries for different machines, and we 
> don't want to
> > use 5.X in production until it is STABLE.
> > 
> > I want to echo the above question. Are there patches available or
> > forthcoming to fix the problems with the e7520?
> > 
> > Thanks,
> > 
> > -
> > Rob Watt
> 
> I think that we can imply from the lack of response on this 
> subject that 4.x
> is not really still supported, since just about all of the 
> new motherboards
> for Intel processors from leaders Dell and Supermicro are 
> based on the 
> 7520. So, ironically, in order to use the newer, faster 
> processors with 
> FreeBSD, you have to use the newer, slower version of the O/S. Yikes!

Considering 4.x kept getting tweaked all the way to 4.10 where 4.10 is
benched even faster than 4.9 in my experiences with the OS, I find it
reasonable to assume that new code that hasn't fully matured and been fully
adopted by mass may be slower. As in your previous post on the subject, I
find it no where near as slow as you have stated. For one who couldn't
figure out how to compile without the witness options and various other
debug stuff into the kernel and base system, it prolly would be slower.
After I took this stuff out of the build my benches were greatly improved,
but alas, not to 4.10 speeds. Given time and mass acceptance of the OS which
will adopt more coders and patches from people not yet using the OS I would
expect it to speed up even more. When you do such a radical swing in
codebases and haven't fully tested them, you should expect it.