[prev in list] [next in list] [prev in thread] [next in thread] 

List:       beowulf
Subject:    Keyboardless/mouseless/videoless MB's and clusters?
From:       goebel goebel () his ! com
Date:       1999-10-21 13:39:14
[Download RAW message or body]


On Tue, 19 Oct 1999, Stanley, Jeremy wrote:

> My main desire for this functionality would be so I would not have to
> purchase video cards and insert them in headless nodes.  The only way I
> see this new hardware addressing the problem is if this interface is
> either integrated into a motherboard or significantly cheaper than a
> simple monochrome display card.  I presume there are other reasons than
> cost+time for wanting this type of interface...  Anyone?
> --

We have a program call VACM <http://www.valinux.com/projects> that is a
low level management tool that you can use on Intel server boards to get
to the BIOS via a serial connection. It supports something called EMP -
emergency management port.

Through VACM you can access the BIOS on individual nodes, and groups of
nodes.  You can craft a BIOS setting by hand, but also can send out group
'style sheets' to multiple machines to change BIOS settings.

The interface will let you power up and down a node; detect intrusion; fan
speed; and you can download logs of low level hardware events. There is a
system snapshot that is being worked on, so if you want to see resource
usage, you can get it through the GUI. There's paging/alerts too.

So the idea is, if you take a rack of machines, one with a serial card,
plug it in and fire that controler node up, remotely you can bring the
rest of the rack online. You can detect that a DRAM in slot 3 of node 11
is ill, and shut down the node, have someone in the co-lo change the DRAM,
and remotely start the machine. You can get a page or email is someone
steps on the cable and downs a machine acidently, etc.

The physical interface to the nodes is out of band. EMP goes through com1,
and we manipulate it through a gtk interface. There is an api for
scripting that is being worked on. We are working towards and inband
management tool via the onboard eepro100 ethernet chip.

Typically on a rack, what we do is redirect EMP for bios level monitoring,
and com1 for serial rediect. In this way we are able to get down to the
BIOS, and get all the way up to the console on each node without having to
leave home.

We would like other people to use it. We are doing development now on the
inband management, the ablity to push out fireware updates, and a
database management front end to monitor the state of the cluster.

It's open sourced, GPL'ed. This is an early version, so don't expect too
much, but it works and is useful.


John Goebel
VA Linux Systems



-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to beowulf-request@beowulf.org

[prev in list] [next in list] [prev in thread] [next in thread]