DELL PE 2850 - problems with SMP and PAE kernels

[Stephane Raimbault]

Hi,

I have been working with a Dell PE 2850 here recently and I've been  
having some weird problems.  I noticed a few previous posts about a  
PE 2850 and some problems, but I think my problems are slightly  
different, however somewhat similar.

I'm setting up my PE 2850 (dual 2.8 xeon, w/ 4G of RAM and a PERC4/dc  
raid card) to be a mysql server.  Right now, imysql is setup as a  
slave to replicate from the master, with the eventual goal of  
replacing the master with this server.  I have ran into a couple  
problems however, while running the SMP or the PAE kernel.  GENERIC  
seems to be okay for me.

I have the port mysql41-server installed WITH_LINUXTHREADS=yes  
BUILD_OPTIMIZED=yes BUILD_STATIC=yes as I do on my other non Dell  
boxes running FreeBSD 4.x and mysql.

I first attempted to run the PAE kernel to take advantage of all  
4Gig's of RAM.  I was pleased with performance, however while my sql  
server was slaving from my master in the middle of the night it  
panic'd.  I wasn't at the console to see what the panic was about...  
so I re-setup my server and started slaving again, this time logging  
my console.  This is what came up when it panic'd again:

panic: lockmgr: thread 0xfffffffe, not exclusive lock holder  
0xc6562c00 unlocking


I decided to try the SMP kernel out at this point as I figured I'm  
hitting some bug in PAE.  I booted into my kernel.old which was an  
SMP build I had done earlier and everything seemed to work fine,  
however I noticed my performance was terrible.  Trying to re-sync  
from the slave seemed to be REALLY slow.  My CPU's seemed mostly idle  
(95 - 99% idle) during the sync, however my system load was 1.0 - 1.5  
instead of 0.33 under PAE.  I couldn't explain what I was seeing...  
so I decided to try GENERIC.

My experience with GENERIC was that my performance returned back to  
how it was with PAE, however the system did not crash while syncing  
from the sql master server.  The server continues to operate and  
remain synced with the master... so all seems well, for the time being.

I was hoping that someone could provide some insight in what I'm  
seeing.  I can run some benchmarks and tests for the time being with  
this machine, as I don't plan on putting this box into production as  
is until at least next week.

I also have the following in my /boot/loader.conf to allow mysql to  
use up to 2Gig's of RAM on this system:

kern.maxdsiz="2147483648"



Looking at some previous posts, I noticed that people were reporting  
some usb interrupt related messages, which I am also seeing, however  
I don't believe this is the cause of my problems since I was running  
the PAE kernel which has nodevice for the usb devices, however I  
still had usbd_enable=YES in my rc.conf.  Would this enable the USB  
devices despite having a nodevice uhci in the kernel file?

Thank you,
Stephane