IOUG Live! 2003 Day #2 (Monday)
by Stephen Andert
Rich presented a couple of awards, including the Chris Wooldridge Award winner for outstanding volunteer this year is Tony Jedlinski, editor of SELECT magazine. He also presented the SELECT Editor's Choice Award to Edward Kosciuszko for his article "Analytical Functions in 8i". These are the kinds of people that make IOUG successful.
Rich then turned the session over to Emery John "EJ" Bodnar, Director of Marketing for Business Critical Systems of HP Americas.
EJ talked about 4 areas that show why he feels HP is the best partner for Oracle professionals. He cited the significant areas where HP & Oracle have entered into partnerships. He also mentioned the areas where Linux will complement HP-UX and HP's Linux initiative. In his opinion, Linux fits best on Network edge right now and is probably not quite ready for mission critical databases especially since accountability better with non-linux vendors. One of the benefits of integrating both Linux and HP is the cost savings and portability that allow you to develop on Linux and deploy on HP.
Another area EJ discussed was the reasons for and the benefits of Itanium. The partnership between Intel & HP was announced in 1994 which shows the level of committment that HP has to the Itanium chipset. He said that Linux with Itanium are disruptive technologies that HP is poised to fully take advantage of.
Finally, EJ discussed RAC and the role HP (and historic Compaq/Digital) played in developing the RAC technology and one was customers can take advantage of it by consolidating onto fewer hosts and in the process save money and provide better availability.
One of the things that I value about this conference is the opportunity to hear different points of view. The "hit ratio" vs. "wait interface" debate has been simmering just below the surface. Today I heard arguments for and against these opinions. To try and focus on the technical points, I'll deviate from my usual habit of identifying the specific presenters and just paraphrasing the points that were made in sessions that I attended today. Any of the presenters or attendees are welcome to add their comments in the talk back section at the end of this article.
Sometimes, it is not possible to identify the cause of a performance problem using aggregate.
The number of times that something was executed does not give any idea how much time that it took to run.
Hit-ratios are not the only tool to use in approaching a tuning problem.
Hit ratios can be proactive early warnings of a major problem.
Very bad queries can create an artificially high hit ratio.
These are some examples of the debate that I've seen so far. I know I've formed some opinions on the matter, but I'll leave it up to other dba's as to how they want to implement these various schools of thought into their database tuning activities.
William Vollenweider gave a great presentation on Key Compressed Indexes. One of the biggest benefits of doing this is the space savings that are possible. Also, in his testing, creating compressed indexes always has run faster than non-compressed.
The afternoon keynote speaker was Andrew Mendelson from Oracle Corp. He presented an update on Oracle 9i. He followed that with success stories from customers with a panel that included Rich Niemiec, Tim Gorman, Jared Still, Kirtikumar Deshpande. They each discussed their experience with 9i and what features were most important to their implementation of 9i.
Mendelson followed that with a discussion of some of the next steps to be included in future releases that will involve better managability built in that will make the database more self-aware, self-tuning and self-maintaining. I will leave it to the reader to decide how soon to expect these features.
Jonathan Lewis presented "How the Cost Based Optimizer works", in which he presented a puzzle. He created 2 tables with identical data in two different ways and then showed how the same query acted differently on each. He then used this puzzle to show how when the CBO comes up with what we think is a non-intuitive plan, it is actually logical based on the data, layout of that data. This was a great presentation that provided attendees with a wealth of tips that can be used to enhance database performance.
A group of us finished the day after the welcome reception by gathering in the lobby. It was fun to put some names from mailing lists to faces. The general feeling was that the group was too large to be able to go anywhere and remain a group so we split up into smaller groups and went in different directions. The group I went with went over to Downtown Disney and did some shopping. Our group the split again as some people wanted to do more exploring and others needed to call it an early night as we dealt with jetlag and conference exhaustion.
This was another great day and I'm looking foreword to tomorrow.
What are your thoughts on the Hit Ratio vs Wait Interface debate?
hit or wait?
There are numeours technical reasons on why the hit ratios are no good. But there is a more important business reason for the IT department why they should go with response time analysis across the tiers: It shows the problem in an obvious way (which tier is consuming the most cpu + wait time). How does the buffer cache hit ratio hbelp you with a network problem? The response time way is repeatable and clear for all parties involved and it will safe a lot of money (faster in finding the symptoms and the root cause across all the tiers).
hit ratios not a hit
The mindless-ness of hit ratios can be trivially demonstrated by going to Connor McDonald's website (www.oracledba.co.uk) where you can download a script to generate ANY hit ratio you like...Proof that its so easy to be misled by a ratio