Oracle Magazine - March/April 2012

and explain their real-world performance techniques.” So three of us put together a daylong seminar, which became the Real World Performance Tour.

The three people involved are myself; Graham Wood, who’s been with Oracle since the early 1980s and is actually the father of Statspack, Automatic Workload Repository, and related tuning technologies; and Andrew Holdsworth, who also has been with Oracle longer than I have. He is the senior director of the Real World Performance group within Oracle and works with customers and prototypes and on fixing systems that are not working in the real world. We get together for a day and talk about data warehousing in the morning; then in the afternoon, we talk about OLTP.

If people are interested in these Real World Performance Tour sessions, we’ll be doing more in 2012. Just look at ioug.org under Events, and you’ll see a list of dates that we’ll be appearing—maybe in a location near you. Oracle Magazine: Tell us more about what goes on at these Real World Performance Tour events.

Kyte: Well, they’re very interactive. We have an audience that’s typically anywhere from 40 to 100 people, and even with 100 people, you’d be surprised at the degree of interaction and the back and forth that goes on. And while we’ve given the seminar more than 15 times now, we’ve never covered the same material in the same way twice. In fact, sometimes we don’t even cover the same set of slides twice in a row because the conversations we’re having take us down different paths.

One of the unusual things about the Real World Performance Tour is that we have a rather large machine with us. We have access to an Oracle Exadata full-rack machine, so we’re able to turn around and prove the points that we’re talking about. We can put numbers behind what we’re saying. So sometimes there’s a lot of unlearning on the part of the attendees during these days. Unlearning something that you’ve learned is true for the last 10 or 20 years can be a very painful process. Oracle Magazine: In so many years of answering questions, are there any that

The best questions are when I have to fire
up SQL*Plus and develop a small test case—
an example—to either prove or disprove a
point. Those are the best to me because
they’re the most fun.

really stand out? Is there a best question or maybe a worst question?

Kyte: There have been so many questions— more than 33,000 on Ask Tom in the last dozen years. So coming up with the best or worst question is not really possible, but let me say this: I usually start the morning by answering some questions. If I can get through 10 questions and I haven’t had to get into the database to answer anything, that’s sort of a bad day for me. That means the questions have been asked and answered previously. I simply answer them by taking the subject of the question and running a search of Ask Tom and saying, “This is what I found when I searched— what did you find?” So that sort of describes the worst questions—the ones that are easily answered because they’ve been answered so many times and people didn’t bother searching.

The best questions are when I have to fire up SQL*Plus and develop a small test case— an example—to either prove or disprove a point. Those are the best to me because they’re the most fun. I have to actually think about what I’m doing when I do the answer. And they’re generally questions that have not been asked before, or not been asked in that way, so they’re sort of new.

Oracle Magazine: What’s the best way to ask a technology question?

Kyte: Step one should be to go ahead and look to see if you can’t solve it yourself really quickly by taking a look at the documentation or actually using the search features on the sites where you’re going to post the question. Because 9 times out of 10, you’ll find that your question has been asked and answered by someone else.

Once you think you have a unique ques-
tion that you can’t answer yourself, when
you post the question—and this goes for
Ask Tom as well as any other forum—ask the
question as if you were asking your mom the
question. Give that level of detail. The people
you’re asking haven’t been sitting staring
at the problem for a day or a week, like you
have. We don’t have all the information you
have. So rather than being very terse and
saying, “How can I get this output from this
data?” explain what the output is. Explain
the logic behind getting that output.