Justin Cormack wrote:
>What do people do when they get very large numbers of serial consoles?
>
>I was wondering if using etherboot-like network drivers (ie no dma) to run
>the console would be any less reliable than a serial console? As I am using
>gigabit ethernet there is often a spare motherboard ethernet (as you
>obvioously dont want to run the main ether like this), and it could be on
>a dedicated network for security (especially if you feed udp packets into
>the console).
>
>It might take a little more setup than serial (ie you might not be able to
>do it before RAM init) I guess, as you have to detect media, and do dhcp
>(though you could run over raw ethernet in broadcast mode).
>
>Is this sensible?
>
>Justin
>
We actually have two console over ethernet systems that we are playinga
round with. One operates completely over raw ethernet using broadcast
and a special Frame ID which is discarded by systems that don't care.
We then run a modified syslog catcher which prepends the MAC Address of
the sender to the message. You can then use typical Syslog filter
mechanisms to get rid of what you don't want to see. You can also do
MAC->IP->Hostname translation.
The other system is a bit more complicated and is more for use in the
kernel. It uses a modified Java telnet client and a UDP protocol. It
opens up a new "tab"bed frame for each host connected and gives you two
way communicaiton. The IP address of the console "server" (where the
Java application is running" is passed via DHCP or as a kernel argument.
The first method is really light-weight, but implicitly one way. It's
usefull for logging and seeing what the hell is wrong with a system
that's not booting. We have versions running in LinuxBIOS and in the
kernel. The second method is more of a KVM replacement, but doesn't
handle things like panics very will (when you TCP/IP stack may go out to
lunch). It's also pretty heavy weight to put in the BIOS.
Some middle ground between the two is probably the most usefull. If
there's general interest, I'll see about including our raw-ethernet
console in our open-source submission to LinuxBIOS (along with Sis635
support). *
-eric
* - We are currently in the process of going through IBM legal to get
approval to open source our modificaitons to LinuxBIOS, it should be
available shortly...however, the hardware platform we are supporting is
a research prototype, not an IBM product so the only really usefull
elements will be the SiS635 support and maybe the Console over Ethernet
support.
