http://lwn.net/Articles/20741/ This is a special feed containing comments posted to the individual LWN article titled "The new NUMA scheduler". hourly 2 http://lwn.net/Articles/21369/rss 2003-01-29T17:12:39+00:00 hbaum The NUMA scheduler patches that have been around since October consisted of two pieces - balance-on-exec, and NUMA-aware-load-balance. The three line patch that Martin wrote replaced the NUMA-aware-load-balance, but still relied on the balance-on-exec patch being in place. So it is not accurate to state that the NUMA scheduler patch which provided equivalent performance to the original NUMA scheduler patch was simply a three line patch. Rather one half of the original NUMA scheduler patch was replaced by a three line patch. The balance-on-exec portion of the NUMA scheduler patch was still required to get equivalent performance.<p>Regarding something like this going in after feature freeze - with Martin's contribution, the NUMA scheduler changes have no impact on non-NUMA systems. The balance-on-exec portion of the patch was always a NUMA only piece of code, and Martin's contribution removed the need for an invasive change to the load balancer. http://lwn.net/Articles/20958/rss 2003-01-23T21:27:52+00:00 mbligh I wrote the &quot;3-line&quot; patch to show how non-invasive a NUMA scheduler could be. It also changes the design of the NUMA scheduler to become a set of &quot;concentric circles&quot; as far as load-balance is concerned - I think this is a better design.<p>IMO, in the post-freeze environment, the only way to proceed is small, well contained patches that obviously don't break anyone else ...<p>M. http://lwn.net/Articles/20849/rss 2003-01-23T03:02:05+00:00 corbet Sorry, I should have been more explicit there. The patch was by Martin Bligh, who has been working with the NUMA scheduler for a while. <a href="http://lwn.net/Articles/20848/">Here's a copy</a> with his explanation of what's going on. http://lwn.net/Articles/20846/rss 2003-01-23T02:54:20+00:00 cpeterso I am a little confused about the origin of the "three-line patch". Was that patch taken <i>from</i> the ongoing NUMA schduler work? Or was it written by an "outsider" who demonstrated the same benefits as the "big" NUMA scheduler, but with far fewer lines of code?