[Beowulf] non-proprietary IPMI card?

[Bogdan Costescu]

On Wed, 29 Nov 2006, Bill Broadley wrote:

> Keep in mind that it's part of the IPMI cost, usually you need a 
> port on a switch, another U, another cat 5 cable, a server 
> motherboard (instead of a desktop), a server CPU (instead of a 
> desktop cpu), a server power supply, etc.

Well, if you don't set up a separate network to be used only for IPMI, 
you don't need some of the items in your list; IPMI access can be 
restricted (password-based) and can use encryption, if you are 
concerned about security.

If you do set up a separate network, IPMI traffic doesn't need at all 
a high speed one - so a 100Mbit switch should be more than enough; 
these days 100Mbit switches even with large amounts of ports (say 24) 
are really cheap.

> Not sure I see the benefit of pre-failure analysis, after all before
> the failure you can use smartd, lmsensors, and related in band data
> collection utilities.

Well, iff you can get lm_sensors to work... you might get some 
temperatures and/or fan speeds. With IPMI is usually simpler, as it 
comes with all the calibration from the manufacturer. (That doesn't 
imply however that all IPMI solutions are equal...).

And because you mention it: I always wondered why IPMI never included 
any kind of data about disks (SMART or maybe only a subset of it). I 
sometimes find a node panic-ed due to a disk failure and I would 
really want to know that a bad sector just couldn't be remapped in the 
last minutes/hours - then I'd directly go to the node with a new disk.

--
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