Again I apologize for the delay in sending out those minutes. I was out of the country for the last 2 weeks and didn't have access to e-mail. I'm back at work now and available if you have any questions. Attendees: George Mann, Bull Ashok Raj, Intel Joel Schopp, IBM Renier Morales, IBM Nathan Lynch, IBM Mary Meredith, OSDL Mark Wong, OSDL Bruce Vessey, Unisys Natalie Protasevich, Unisys Alex Vasilevsky, Virtual Iron David Lively, Virtual Iron Martine Silbermann, HP * Hotplug BOF at LWE. There were about 15 to 20 attendees, a few developers but mostly vendors newly interested in the hotplug market. Since there were only 2 time slots allocated for all the BOFs and since the "Virtualization" BOF was scheduled at the same time, attendance was affected. AR: Mary will contact the organizers and suggest to allow for more time slots for BOFs. * Hotplug for Virtualization. Alex Vasilevsky and David Lively from Virtual Iron are new to the SIG and I asked them to tell us about their product and which part of the hotplug code they currently use. Alex explained how they aggregate multiple x86 type base servers to create a single hw resource that is then used to support multiple virtual machines. They use the hotplug CPU code with some minor changes that will be submitted to mainline. For the memory hotplug they use a node oriented NUMA like approach, some difficulties were mentioned in the remove memory implementation. For platform management (to add/remove components) they use their own "home grown" application. * Patch submissions for CPU hotplug Ashok submitted 3 patches needed to give the offlined CPU back from the OS to the SAL in ia64 architecture. Natalie submitted for review to a small set of people her changes to the CPU hotplug code for ia32 support. * Hardware update. The 4-way system is still not booting, the reason is currently unknown, OSDL is putting an engineer to work on it. * Regression testing for CPU hotplug. Discussion around what should be tested to make sure that CPU offlining works correctly. For example, set the affinity of a task to the CPU that we want to offline, after offlining make sure the affinity of the task has been changed to an online CPU. Also checking that interrupts are handled properly and not lost. Suggestions were made to move packets from outside the box to check if irqs are rerouted properly. AR: Mark to create a list of checks that could be done to verify proper execution of CPU offlining without disruption in the system. Can we trust tools like SAR and TOP? TOP currently crashes when a CPU is taken offline. Virtual Iron has a patch to fix that problem. AR: Virtual Iron will share their patch with this mailing list and eventually submit it to the community. * OpenHPI update. There's a new wiki website hosted by OSDL that Reiner mentioned will be used to generate up-to-date documentation to support the latest release of OpenHPI. The doc currently found in openhpi.sourceforge.net is outdated. The wiki site for the new doc is http://wiki.openhpi.org/OpenHPIWiki The next OpenHPI release version 2.1.0 is expected to be published in June (mostly bug fixes). * Hotplug testing. Through the discussions we've had about what needs to be tested we came to the conclusion that it would be helpful to have a site to keep track of all the different options, configurations and edge cases that we should test. AR: Mary created a wiki site to allow people to submit their thoughts on the subject of testing (this can be extended to include what features should be added that people don't have currently the time or resources to implement). * Program Management needs. Simple things like keeping track of what AR/by whom are done, what patches are submitted, what those patches included and what their current state is (under review, accepted,...) and so on would be helpful information for the SIG to keep track of its accomplishment and to highlight problems if/when they occur. Thanks for your participation. Martine J. Silbermann
