Problem with mpt(4) and Infortrend RAID

Rong-En Fan rafan at csie.org
Sat Feb 19 02:12:45 PST 2005


On Sat, Feb 19, 2005 at 05:45:12PM +0800, Xin LI wrote:
> Would you please try:
> camcontrol tags da1
> 
> To see how many tags are actually allowed for the device, at the host
> side?  It comes to my mind that:

I have already done -N 32, I remembered that original is 25x.

> ÔÚ 2005-02-19ÁùµÄ 17:23 +0800£¬Rong-En FanдµÀ£º
> >  Maximum Queued I/O Count -  256                                
> >  LUNs per Host SCSI ID - 8                                      
> >  Max Number of Concurrent Host-LUN Connection - Def(4)          
> >  Number of Tags Reserved for each Host-LUN Connection - Def(32) 
> 
> Here:  You have allowed only 32, and if the host has exceeded the
> limitation, then something strange *may* happen.  My bet is to decrease

I just found the Infortrend's manual, it says that if
hos exceeded this limit, the controller will not response
and host *might* reduce the # later.  Hmm, it seems match
my situation.

> the tags allowed at host side, to something like 32:
> 
> camcontrol tags da1 -N 32.
> 
> It's also possible to modify cam_xpt.c to set the default number.

Hmm, seems ok for me. My system's disk is ips(4) which does not
use CAM layer. So, I can put camcontrol to rc.

> > I'm not that similar with SCSI, if it possible some parameters
> > here need to tune?
> 
> If you have a spare box, try to decrease the tag depth :-)  Please be
> sure to decrease tag depth on every device.

I got no spare box here (to be precise, I can not find another
320 scsi card). So I directly decreast the tag depth on the
main server. Hope there is no more timeout and thus I can
have a good sleep.

Thanks,
Rong-En Fan


More information about the freebsd-scsi mailing list