SQL PASS Summit 2010 – Keynote Day 3

Today is the last official day of the SQL PASS 2010 summit, which also means the last day for live blogging during the keynote.  This has been a lot of fun, and I’m looking forward to doing it again next year.

Today’s ceremonies will include a keynote by Dr. David Dewitt.  Before that, PASS board member Rick Heigis gives the daily briefing on the current goings-on at PASS, including a review of the most recently elected board members and a recognition of outgoing board member Lynda Rab.

Next up, Rick talks about the upcoming SQL Rally, announcing the winners of the community vote for the preconference sessions.  You can register now for preconference sessions delivered by the Pragmatic Works crew, Plamen Ratchev, Kevin Kline, and Grant Fritchey.  Registration is now open for the Rally and the preconferences for a discounted bundle rate.  We’re already looking forward to the next PASS summit.  Registration is available now for $995, or you can attend the summit plus two full days of preconference material for $1295.


Dr. David Dewitt is up next to discuss SQL query optimization.  Instantly he’s a crowd favorite, bringing in a few timely puns before getting started on the main content.  He begins by giving a high level overview of the query optimizer and describing why the building of an effective execution plan still remains challenging after 30+ years of RDBMS development.  To demonstrate his theory, he’s using an analogy similar to the Nexflix model which should be at least somewhat familiar to most database professionals.  He gets rolling quickly, describing two different query plans and reviewing the process through which the query optimizer selects the best plan.  By the way, the entire slide deck for this presentation will be available on the Microsoft Jim Gray Systems Lab group Facebook page.  There was a mention of a Q&A later through which attendees can submit questions to Dr. Dewitt – email AskDrDeWitt@sqlpass.org to get your question added to the queue.

This is really good stuff, and he’s moving quite fast through the material.  My only concern is that there’s probably a significant number of people in this room who are completely lost.  This material is relatively specific and deep for a keynote; this topic could have easily been moved to a spotlight session (or two consecutive community sessions).  Nonetheless, there are some excellent concepts here, and I expect that everyone is getting at least something out of this.  Folks that spend their careers doing performance tuning are drooling on themselves right now.

Dr. Dewitt has mentioned a couple of times that query optimization has significant room for error.  It’s a value-based decision, and as smart as the optimizer is, there will be some optimization choices that are simply wrong.  How does this happen?  Simply put, lots of variables.  Imagine your drive to work every day (note that this is my analogy, not part of the presentation).  You have two possible routes, A and B, and you know from having taken both paths that route B is usually faster than A.  On your way, you hear on the radio that route B is partially blocked by a traffic jam, so you opt for route A.  Is that better?  Maybe, but what if everyone heard the same report and jumped onto route A, causing a bottleneck on that path.  What if route A is also blocked by an accident just before you arrive?  What if the accident on B is cleared up and traffic starts flowing normally after you change routes?  A trivial example, yes, but you can see how just a few variables can have a significant effect on the efficiency of a given path.

Dr. Dewitt said that he job of the QO is NOT to find the best plan: It’s to find a good plan fast.  It’s the Nike approach: Just Do It.  What if the query optimizer gets it wrong?  Yes, it happens.  But the logic remains that a plan that isn’t the most optimal is still a pretty good plan.

Statistics are discussed throughout the presentation.  Dr. Dewitt points out that current statistics are essential for building a good query plan.  “If you don’t update your statistics, don’t blame us”, says he.  Good advice, since I suspect that a lot of folks don’t spent a lot of time focusing on this (and I’m among them).

At the end of a brain-busting presentation, many in the crowd (including everyone at the blogging table) gave Dr. Dewitt a well deserved standing ovation.  Dr. Dewitt now takes questions submitted via e-mail during the address.  I really like this interactive approach – it helps to engage the audience and keep the content grounded and relevant.


Today was the last keynote of the conference, and by far it was the most informative.  We had over an hour of pure technical information, which, judging by the feedback in the room and on blogs and Twitter, is strongly preferred to the marketing-laden message delivered yesterday.  Hopefully we’ll see more of the former in coming years.  Look for a wrap-up post in the next couple of days, where I’ll review what I learned and outline my accomplishments at this year’s Summit.

About the Author

Tim Mitchell
Tim Mitchell is a data architect and consultant who specializes in getting rid of data pain points. Need help with data warehousing, ETL, reporting, or training? If so, contact Tim for a no-obligation 30-minute chat.

Be the first to comment on "SQL PASS Summit 2010 – Keynote Day 3"

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.