Oct 14: OOW 2009 - Tuesday
Another good day although it didn't go quite as I planned it, but that's part of the fun of a good conference. Should I mention at this point that I have a cold? I certainly noticed it this morning 
I got moving eventually and my first session was "The Life of a Query" with Jonah Harris. I attended this because I've seen Jonah's previous work via his blog and Koert the Dude Guy really rates him. I trust Koert's judgment on most things, not just Oracle stuff.
Jonah's presentation was about the low-level round trip that happens when you submit a SQL statement to the server. There was a good crowd - room almost full - although I suspect it was the room that Jason Arneil blogged about, so I made sure I nabbed a seat on the same side of the room as the screen. What surprised me about Jonah's presentation was that, despite the technical content and his obvious love of low-level detail, his slides and his presentation skills were top-notch. i.e. Although this was a technical presentation, he was very easy to listen to.
In the end, I bailed out half-way through, but that was entirely down to feeling pretty unwell and because it was a presentation that demanded attention. Once I knew I could look at the slides later, I decided it was more expedient to pick up some lunch!
Next I went to Oracle Closed World, Mogens Norgaard's latest guerilla conference organisation. He's booked a room with beer and food and invited some of the best speakers to give fairly informal presentations to a small audience. Today was Jonathan Lewis - but with the same damn presentation as yesterday! Except it wasn't really. He went into much more depth and spoke for longer but, even though there was slightly less reticence to answer questions from this audience, it was still there
In the end I skipped the 11g Optimiser presentation, knowing that there's an associated white paper that I can read later and, after food had been served, Mogens encouraged Kevin Closson to get up front for a very adhoc Q & A session. The relaxed atmosphere meant that Kevin was even more forthright and entertaining than normal and it was good fun and I learnt a few new things. But I had to leave pretty soon after he'd finished to try and squeeze in my last couple of presentations of the day from the main conference agenda.
Andrew Holdsworth's "Current Trends in Real-World Database Performance" was his usual annual review of, in many cases, the same old performance problems his group see although with a slightly different slant.
The new slant was that, in a world where his team are seeing high performance configurations executing 30-40,000 SQL statements per second. What happens if something momentarily goes wrong? What chaos might ensue as the system tries to make that time back again .... So he spent time talking about what I'd call performance management.
Sometimes you can spot themes at a conference and I always think those themes reflect the problems people are seeing in the real world. This year, I've been in many presentations that have mentioned systems that are hopelessly over-loaded or under-configured (except the Exadata ones, obviously! LOL). It's something I seem to have discussed frequently with people at work over the past few months. Why, oh why, do people seem to think it's acceptable to keep pushing more workload through the system as long as they're not pinning CPU utilisation at 100% for long periods of time. The utilisation figures I'm seeing are *way* too high and then people wonder why they have intermittent but wide-ranging performance chaos?
Cary Millsap talked about this at length on Monday, but Andrew threw in another number for OLTP or mixed workload systems or those with variable arrival rates where users expect consistent response times - something like 60-65% utilisation? But let's not focus on a single number, which will change as the number of CPUs increases, just look at how low that number is. He stated that he thought that CPU over-utlisation was the most common mistake his group sees, outside of Poor SQL (and, of course, the two are often going to be related). Whatever the appropriate utilisation figure might be for your configuration, he suggested (and hallelujah to this!) stop using average CPU utlisation figures, because they hide the actual response times people are experiencing! (However, I'm beggining to suspect I'll die before that message is taken on board fully by a majority of systems management professionals.)
For my next call on what I should attend, I again decided that the In-Memory Parallel Query was something I could pick up via the White Paper and plumped for the Terabyte hour, again from the Real World Performance Group. I'd expected Greg Rahn to be part of the presentation team, but it was Andrew Holdsworth again, along with Mike Hallas .... running at the same time as a Tom Kyte presentation! So it was a little quiet in there.
This presentation looked brilliant - a demonstration of loading 1 Terabyte of user data using compression, partitioning, a sensible tablespace layout, a DBFS staging area on the Exadata box etc, etc. The largest LINE_ITEMS table had just shy of 7 billion rows. I like the way they've set this up as a decent representation of real world requirements which means that when the slides become available, you should be able to use it as a starting point for your own implementations. Unfortunately, a call cropped up and I had to leave early, but I'll be checking those slides out carefully later.
There'll probably be another Tuesday post to come, but for now I'm going to go and see Tom Kyte's Q & A session and then head over to Oracle Closed World.
Disclosure: I'm attending this year's Openworld Conference at the invitation of the Oracle ACE Director program, which is paying my travel and accommodation expenses. The time off work is my own choice.
I got moving eventually and my first session was "The Life of a Query" with Jonah Harris. I attended this because I've seen Jonah's previous work via his blog and Koert the Dude Guy really rates him. I trust Koert's judgment on most things, not just Oracle stuff.
Jonah's presentation was about the low-level round trip that happens when you submit a SQL statement to the server. There was a good crowd - room almost full - although I suspect it was the room that Jason Arneil blogged about, so I made sure I nabbed a seat on the same side of the room as the screen. What surprised me about Jonah's presentation was that, despite the technical content and his obvious love of low-level detail, his slides and his presentation skills were top-notch. i.e. Although this was a technical presentation, he was very easy to listen to.
In the end, I bailed out half-way through, but that was entirely down to feeling pretty unwell and because it was a presentation that demanded attention. Once I knew I could look at the slides later, I decided it was more expedient to pick up some lunch!
Next I went to Oracle Closed World, Mogens Norgaard's latest guerilla conference organisation. He's booked a room with beer and food and invited some of the best speakers to give fairly informal presentations to a small audience. Today was Jonathan Lewis - but with the same damn presentation as yesterday! Except it wasn't really. He went into much more depth and spoke for longer but, even though there was slightly less reticence to answer questions from this audience, it was still there
In the end I skipped the 11g Optimiser presentation, knowing that there's an associated white paper that I can read later and, after food had been served, Mogens encouraged Kevin Closson to get up front for a very adhoc Q & A session. The relaxed atmosphere meant that Kevin was even more forthright and entertaining than normal and it was good fun and I learnt a few new things. But I had to leave pretty soon after he'd finished to try and squeeze in my last couple of presentations of the day from the main conference agenda.
Andrew Holdsworth's "Current Trends in Real-World Database Performance" was his usual annual review of, in many cases, the same old performance problems his group see although with a slightly different slant.
The new slant was that, in a world where his team are seeing high performance configurations executing 30-40,000 SQL statements per second. What happens if something momentarily goes wrong? What chaos might ensue as the system tries to make that time back again .... So he spent time talking about what I'd call performance management.
Sometimes you can spot themes at a conference and I always think those themes reflect the problems people are seeing in the real world. This year, I've been in many presentations that have mentioned systems that are hopelessly over-loaded or under-configured (except the Exadata ones, obviously! LOL). It's something I seem to have discussed frequently with people at work over the past few months. Why, oh why, do people seem to think it's acceptable to keep pushing more workload through the system as long as they're not pinning CPU utilisation at 100% for long periods of time. The utilisation figures I'm seeing are *way* too high and then people wonder why they have intermittent but wide-ranging performance chaos?
Cary Millsap talked about this at length on Monday, but Andrew threw in another number for OLTP or mixed workload systems or those with variable arrival rates where users expect consistent response times - something like 60-65% utilisation? But let's not focus on a single number, which will change as the number of CPUs increases, just look at how low that number is. He stated that he thought that CPU over-utlisation was the most common mistake his group sees, outside of Poor SQL (and, of course, the two are often going to be related). Whatever the appropriate utilisation figure might be for your configuration, he suggested (and hallelujah to this!) stop using average CPU utlisation figures, because they hide the actual response times people are experiencing! (However, I'm beggining to suspect I'll die before that message is taken on board fully by a majority of systems management professionals.)
For my next call on what I should attend, I again decided that the In-Memory Parallel Query was something I could pick up via the White Paper and plumped for the Terabyte hour, again from the Real World Performance Group. I'd expected Greg Rahn to be part of the presentation team, but it was Andrew Holdsworth again, along with Mike Hallas .... running at the same time as a Tom Kyte presentation! So it was a little quiet in there.
This presentation looked brilliant - a demonstration of loading 1 Terabyte of user data using compression, partitioning, a sensible tablespace layout, a DBFS staging area on the Exadata box etc, etc. The largest LINE_ITEMS table had just shy of 7 billion rows. I like the way they've set this up as a decent representation of real world requirements which means that when the slides become available, you should be able to use it as a starting point for your own implementations. Unfortunately, a call cropped up and I had to leave early, but I'll be checking those slides out carefully later.
There'll probably be another Tuesday post to come, but for now I'm going to go and see Tom Kyte's Q & A session and then head over to Oracle Closed World.
Disclosure: I'm attending this year's Openworld Conference at the invitation of the Oracle ACE Director program, which is paying my travel and accommodation expenses. The time off work is my own choice.
Tracked: Oct 14, 22:51