[Freeipmi-users] [RFC] FreeIPMI in heterogeneous environments/configurations

[Al Chu]

Howdy everyone,

I wanted to get people's inputs on this.

The hostranged support in FreeIPMI has allowed parallel communication to
multiple remote hosts:

ipmi-sensors -h node[0-4] -u XXX -p XXX

The problem is that this will only work in homogeneous environments
where everything is identical.  Under most scenarios, this is a
perfectly acceptable.  I think it is reasonable for users to configure
identical usernames/passwords/etc. on those machines they want
to communicate with in parallel.

However, there is a big issue when vendor motherboards are not IPMI
compliant and a user needs to specify one of the workarounds listed in
the manpage.  This is something out of a user's control.  

Another question is, do we need heterogenous support for hostrange
options?  For some tools, it doesn't seem to make sense.  For example,
ipmi-sensors will be outputting different sensors from different
motherboards.  However, ipmipower just outputs power on vs. off, so it
is more agnostic to the motherboard it communicates with.  It may also
be difficult to continually remember which machines need what
workarounds or have alternate configurations.

My initial thought on how to deal with this (this is high level thought,
not much detail yet) is to support a config file /etc/freeipmi.conf
where users could configure specific nodes to use different options to
different hosts.  For example, it could say:

workaround endianseq nodes[0-4]
workaround intel20 nodes[5-8]

The above would indicate default values for the specified nodes when a
user runs ipmi-sensors, ipmipower, etc.

This config file would presumably replace ipmipower.conf and
ipmiconsole.conf and provide tool specific options as well. i.e.

ipmiconsole_dont_steal true nodes[0-8]

What are people's thoughts?

Al

-- 
Albert Chu
address@hidden
925-422-5311
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory