[Beowulf] non-proprietary IPMI card?

Bogdan Costescu Bogdan.Costescu at iwr.uni-heidelberg.de
Wed Nov 29 06:40:35 PST 2006


On Tue, 28 Nov 2006, Bill Broadley wrote:

> I've heard of folks successfully getting the tyan card to work:

On what mainboard ? I have some of these IPMI cards here and they work 
only after they are flashed with a file specific to the mainboard they 
are mounted on. This presumably defines how sensors are connected, 
calibration/scaling factors, etc. It's conceivable that a foreign 
mainboard could be made to work this way too, but someone would need 
to spend the effort of defining these data and I suspect Tyan won't 
lend any hand to help ;-)

> Does anyone know if IPMI can "share" a GigE port?  Or does it 
> usually require a dedicated port on the host?

I have mainboards from Tyan and Supermicro (among others), both with 
IPMI.

The Tyan mainboards, using Broadcom NICs, require assigning to the 
IPMI NIC a different MAC than the real NIC - very similar in concept 
to a VLAN: the hardware is the same, but they are logically different, 
packets sent to the different MACs will be received by the different 
entities (BMC and Linux NIC respectively). The switch sees 2 MACs on 
the NIC and the communication is clearly separated at ISO Level 2.

The Supermicro mainboards, using Intel NICs, require assigning to the 
IPMI NIC the same MAC as the real NIC - and if you make the mistake of 
assigning a different MAC, the IPMI BMC will not be reachable. I think 
that the packets are filtered and everything that looks like an IPMI 
packet is hijacked by the IPMI part, the NIC as Linux sees it doesn't 
get to even notice this packet. The switch sees obviously only one MAC 
assigned to the NIC.

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu at IWR.Uni-Heidelberg.De



More information about the Beowulf mailing list