Hi,
On Mon, 27 Mar 2006, Albert Chu wrote:
Hi Paul,
I'd enabled hardware flow control (as Steffen Grunewald had described
on the ipmitool list, thanks Steffen!)
I can't remember the exact kernel internals logic/reason, but I believe
if hardware flow control is turned on, the kernel will spin(?) (atleast
eat up a lot of cycles) waiting for the ability to dump data out the
console. We've only seen this with normal serial connections, but it
wouldn't surprise me if the same applied for SOL.
While re-running this, via top I do see kipmi0 using ~1% - 7% (yes, not a
good metric ;-)), but that's about it. The machine is otherwise as snappy
as usual.
If the above case happend, it would suggest ipmitool wasn't capable of
handling the SOL traffic quickly enough. That's unlikely. But as you
said, perhaps things haven't been stress tested.
I'm not sure... one thing about these BMC's is that they use what I think
to be ridiculously small packets (with my test, 144B)... in fact a simple
test shows that they'll only respond to pings of <= 302B.
A few UDP packets lost
here and there on the network and ipmitool not ack-ing SOL packets
outside of a certain sequence number range could probably cause it.
(Note, I don't know the internal logic of ipmitool for this example, its
just an idea.)
Interesting.
Has anyone else done any "stress" testing and seen a BMC just fall
down?
Never to the degree that you've stated. Although there are situations I
can imagine where it can happen. For example, if the maximum number of
Lan sessions were alive and connected to the BMC (presumably in
background/sleeping processes you weren't aware of it), then the BMC
could be out of resources and may not respond to additional traffic.
But that's just a guess.
OK, perhaps Duncan can chime in? ;-)
A fun black-magic trick I found with some "locked up" BMCs was to
simultaneously ping the node w/ rmcp and ipmi at the same time. For
some magical reason, this "unlocked" some BMCs for me. I use 'rmcpping'
and 'ipmiping' which are in the FreeIPMI project.
nifty! I'll see what happens, if I'm able to reproduce...
Thanks!
Paul
-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642
_______________________________________________
Openipmi-developer mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openipmi-developer
