Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 18 Jul 2007 09:33:45 -0400
From:      Tony Holmes <tony@crosswinds.net>
To:        freebsd-current@freebsd.org
Subject:   7.0/6.2, 4GB and AHC
Message-ID:  <20070718133345.GA82366@crosswinds.net>

next in thread | raw e-mail | index | archive | help
I had a very interesting couple of days trying to figure out why 6.2-STABLE
and 7.0-CURRENT, both amd64,  were giving me random hangs, panics etc.

My system:

Asus M2NPV-VM
4GB Ram
2 x Adaptec 19160 controllers
14 x 33GB Seagate 10k rpm scsi in a disk shelf (2 channel)
2 x 80GB Seagate SATA (geom mirror, system disks)

I had the system installed with 6.2 and upgraded to 7.0- (for the new nfe
driver for the onboard NIC) and added the Adaptecs and disk. I began to play
with geom to set them up as a RAID 10 system for a database. Same configuration
on FreeBSD, solaris and other systems resulted in blazing fast performance.

Set up the disks, fired up a basic bonnie test and... chug chug and eventual
random panic. Okay, maybe I messed up the geom config since I was tinkering
around. Reboot, wait for sync to finish.... 20 hours later it was done - way
too slow. Checked termination, cables, power, etc. All good.

Fire up bonnie... first iteration showed approx 8MBps write approx 10MBps read
was horrendous. Decided to attempt a fresh reinstall.

Load up 6.2R amd64 cd - panic on startup just after SCSI probe delay. Tried
once more and same thing. Okay, try 7.0 amd64.

Divide by 0 error right after SCSI probe - and that's when the thought struck
me - 4GB ram... 

Pulled out 2GB and viola, 7.0 installed easily. Configured the Raid 10 and got
the nice 275MBps read and 150MBps (ballpark) benchmark numbers I was expecting.

I know that amd64 supports 4gb+ (I have 2 others with SATA only that are
running flawlessly).

So I am attempting to determine the cause of the failures.

The adaptecs are 32bit with older bios (2.57 and 3.10). They are in the 2
32bit pci slots of the M2NPV motherboard.

I would have thought FreeBSD would have knocked a memory hole in the 3.5-4gb
range to accomodate the device mappings.

Does anyone have any explanations/pointers (I did search and attempt to RTFM
with not much luck) about this?

-- 
Tony Holmes

Ph: (416) 993-1219

Founder and Senior Systems Architect
Crosswinds Internet Communications Inc.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070718133345.GA82366>