Re: Interrupt routing and friends ...

[Bill Hacker <wbh@xxxxxxxxxxxxx>]

Justin C. Sherrill wrote:
> On Sat, November 8, 2008 4:32 pm, Joerg Sonnenberger wrote:
> > On Sun, Nov 09, 2008 at 04:56:21AM +0800, Bill Hacker wrote:
> > > It might be worth a look at recent OpenBSD releases.
> > I have a very different opinion of that code.
>
> Where would you recommend looking?

For starters, what put us off FreeBSD:

- taking the all-too-ubiquitous Intel ICH7 thru 9 family - (Tyan, Asus, 
GigaByte Core-D/Core-2 MB), FreeBSD had been erratic to the point of 
requiring either staying with older releases AND NOT 'STABLE', OR 
disconnecting from that onboard chipset and moving the drives to 
SiliconImage add-on controllers. Both PATA and SATA drives.

BIOS setting to 'legacy mode' helped, but was not enough to prevent 
periodic I/O madness.

About a year+ ago that was 'temporarily' traced to an overly 
enthusiastic mfc'ing from 8-CURRENT to 6.2 STABLE.

- Likewise a period where there were problems introduced w/r Broadcom 
GigE NIC chipsets that had not been problematic in earlier versions.

- parallel port drivers that did not work were not on our radar 
(rackmount servers..)

What let us use OpenBSD as a replacement:

Same or same-rev MB OTOH have had *zero* issues with PATA/STATA or NIC's 
under OpenBSD - 4.2 thru 4.4, including snapshot 'beta' builds at 
roughly 2-month intervals.

Newer MB, and several laptops, old and new,  have held to that 
trouble-free pattern.

CAVEAT:

- Some, not all, of the FreeBSD installs use atacontrol and/or GMIRROR 
RAID1. Only in the last 20 days have we even tested OpenBSD 'softraid', 
and it is not ready for production use. Largely our OpenBSD has been run 
without any software RAID.

On single drives, OpenBSD also unixbench'ed and bonnie++ at roughly 25% 
slower than FreeBSD on same hardware.

There are, of course, a legion of other issues, but I wasn't intending 
DragonFly adopt OpenBSD as a model.

Just that someone who knows what he is looking at should have a look at 
the way they've gone about coding that area vs the way FreeBSD has done.

The Intel IHC drivers and the trail of their plumbing is where I would 
look first. They are about as common as housefly turd.

Regards,

Bill