[BBLISA] 3ware RAID controllers & 1U servers

[Tim Pierce]

Hi all -

I am bringing up some new 1U servers, running FreeBSD 6.1, that have
3Ware 9550SX RAID controllers.  These controllers have battery backup
mounted on the card.  I'm relatively new to system administration on
this scale and have a couple of questions about whether my experience
here is typical for 1U machines with cards like these:

1. Yesterday we got syslog warnings that the battery temperature was
   too high.

   The 3ware cards mount to the mainboard by means of a daughtercard,
   and the construction of the cards means that the Battery Backup Unit
   is mounted on the underside of the 3ware card.  That puts the battery
   between the controller card and the motherboard.

   The 3ware manual recommends against installing a PCI card next to the
   battery because it restricts airflow.  However, I can't figure out
   any way to install this controller in a 1U box without sandwiching
   the battery between the controller and the mainboard.

   Has anyone had experience with 3ware controllers with battery backup
   in 1U boxes?  Is this fundamentally an unwise idea or is it worrying
   about nothing?

2. Yesterday before leaving work I ran a battery capacity test (by running
   tw_cli and entering "/c0/u0 bbu test").  The test completed without any
   apparent incident, but immediately afterward we received a couple hundred
   of these messages in syslog:

   Jun 28 06:44:50  kernel: twa0: WARNING: (0x04: 0x0045): Battery voltage is low: 
   Jun 28 06:44:50  kernel: twa0: ERROR: (0x04: 0x0047): Battery voltage is too low: 

   followed by a couple hundred of these:

   Jun 28 06:45:22  kernel: twa0: WARNING: (0x04: 0x0045): Battery voltage is low: 
   Jun 28 06:45:22  kernel: twa0: INFO: (0x04: 0x0044): Battery voltage is normal: 

   The 3ware manual
   (http://www.3ware.com/support/UserDocs/3ware9590SEUsrGuide.pdf) says
   that because a battery capacity test involves completely discharging
   the battery and then completely recharging it, the controller will
   report a low battery voltage during the course of the test and will
   return to normal status.  So I am guessing that these messages were
   generated over the 12 hours of the test and were cached somewhere
   (by syslog? by the controller?) before being written to disk, which
   is why they all appeared in a big clump at the end.

   Does that sound right, or should I be concerned about the battery
   misbehaving?

Thanks in advance,
--twp