Eric Van Hensbergen
ericvh at gmail.com
Mon Jul 25 16:21:07 EDT 2011
according to their previous tests, its not a problem up to four racks
-- however, it is a perceived bottleneck for anything more than that.
>From previous discussions, its conceivably not a problem to move it to
IO nodes and to otherwise federate it for scalability -- its just code
With topics such as that in mind, it may not be a bad idea to map out
kittyhawk milestones for the next year (or two) next week at the
meeting so we can start knocking them down as milestones (and maybe
report them as new milestones to Lucy?)
On Mon, Jul 25, 2011 at 2:51 PM, ron minnich <rminnich at gmail.com> wrote:
> On Mon, Jul 25, 2011 at 12:48 PM, Eric Van Hensbergen <ericvh at gmail.com> wrote:
>> On Mon, Jul 25, 2011 at 2:46 PM, ron minnich <rminnich at gmail.com> wrote:
>>> On Mon, Jul 25, 2011 at 12:07 PM, Eric Van Hensbergen <ericvh at gmail.com> wrote:
>>>> Jonathan can correct me if I'm wrong, but IIRC:
>>>> kittyhawk boots a khctl kernel on one of the compute nodes when it starts up.
>>>> khdo ssh's into that system to perform commands which interact with the other
>>> whoa! interesting! So compute node console is wired to a torus vc?
>> Again, Jonathan can correct me if I'm wrong, but for some reason I
>> think it might still be tree, particularly since console is often used
>> as a broadcast medium.
> that would be tricky, when you get to more than 64 nodes. I wonder if
> that's why it's on a compute node?
More information about the Kittyhawk