To: Kanoj Sarcar cc: andi@suse.de, andrea@suse.de, aono@ed2.com1.fc.nec.co.jp, beckman@turbolabs.com, bjorn_helgaas@hp.com, Hubertus Franke/Watson/IBM@IBMUS, Jerry.Harrow@Compaq.com, jwright@engr.sgi.com, kanoj@engr.sgi.com, kanoj@google.engr.sgi.com (Kanoj Sarcar), kumon@flab.fujitsu.co.jp, norton@mclinux.com, suganuma@hpc.bs1.fc.nec.co.jp, sunil.saxena@intel.com, tbutler@hi.com, woodman@missioncriticallinux.com Date: 03/29/01 08:14 PM From: Paul McKenney/Beaverton/IBM@IBMUS Subject: Re: your mail > I remember Paul asking how IRIX handles striping data across several > nodes for shared memory programs. In the specific case of Oracle, > there is normally an initializer thread that allocates pages, so > this thread has to be numa aware and make the right policy calls > to do striping. Thank you for the info! > An alternative way is to declare policies will affect underlying > objects (the shm segment), which works quite well for cooperating > programs sharing well defined resources. Unfortunately, this will > not work well in case of random programs using dso text for example. > Although you can claim that this case should be mitigated by > replication. > > Unfortunately, another bullet item that I didn't get to discuss. > Look for "Policies on objects or address ranges/spaces". Your presentation certainly did generate quite a bit of discussion, which is to be expected, given your long experience with and many contributions to both NUMA and Linux. I would have liked to have let the discussion go on longer, but it was important for the other people to have a chance to describe their work. Thanx, Paul