On May 15, 2008, at 11:31 AM, Matt Hogstrom wrote:
Well, you can't donate to 'Geronimo' per se only to the ASF. I
wasn't involved directly in this, but I believe the heritage for
the machines is AMD -> IBM. IBM people used the machines for their
geronimo work (including GBuild).
I worked with AMD to acquire these systems. When we got the systems
from AMD they agreed to provide them for the use of the GBuild
project and were never in a corporate owner's asset list. I
provided AMD with the ship to address (which I think was David
Blevins) so continuing to use them for the Geronimo is totally in
line with what the original intention was.
I think the machines would be fine except that they are now about 4-
years old and had some issues that Jason struggled with so they may
not be highly dependable. As far as Lights Out Management I suspect
that the this means that we would likely plug them into a remote
power management unit so they can be power cycled remotely and not
require any manual intervention. I don't know if these machines
would qualify for that kind of support. We'll need to investigate
that.
Is someone tracking them down and do we have a current inventory on
what they are?
As far as machine requirements here is my input. Given that TCK is
largely single threaded a quad-core system would be fine. Running a
hypervisor like XEN would make the most sense. I suggest 4GB per
server instance and with a Quad core that would be 16GB. We can go
lower on memory but we'll since TCK runs servers, adjunct Java
processes, etc we'll grow into the extra head room. So, that said,
we should acquire a Quad-Core 16GB system.
I have some SuperMicro X7DB8+ motherboards (2) I can donate. I have
processors as well (2.66Ghz dual core or quad core). To make these
whole we'll need cases, heat sinks and memory. If we figure out
that this will save us some dough I'd be happy to build the systems
and prepare them to send to Infra. I've been busy with other stuff
but I do have time and resources to help out here.
Should I work up a list of required pieces and approximate prices to
acquire and ship to the remote locations?
Yes, I'm certainly assuming (hoping) that we'll be able to run
multiple images on a single box. We should check to see what sort of
heap sizes we're allocating to the JVM processes during our TCK runs.
To help with calculating guidelines for memory per core.
We also need to discuss with ASF Infra specifics about running XEN (or
similar) hypervisor. We need to be aware of any concerns they might
have...
--kevan
