Thanks for the detailed information on those trace flags. I also eagerly await the blog post.

In an ideal world I would definitely be running replays on our lab environment to observe the effects of these different options. I must say, I am.... suspicious.... of NUMA.

In the real world... well, we'll see if I can get it prioritized. I've already been drawn off in other directions because the server is running "relatively OK" on RTM.

Here's a picture of PLE by node, and free memory (orange line) on RTM. Ignore the giant free memory spike, that was me allocating 10 more Gig to SQL.

In my opinion this is still a pretty wild ride, but it's definitely better than it looked under SP1.