Nov 15: OOW - Day 4 - Performance Day
Yesterday was the best day so far, particularly the morning. To start with, I was feeling a little more human (although still out of bed at 4am). I managed to squeeze in breakfast with APC, three conference sessions and a lovely burger for lunch 
First was Jonathan Schwartz's keynote. You'll have read plenty about that elsewhere, but it was notable for Dell announcing Solaris support and yet another virtualisation announcement. I realise this is a hot topic for most businesses at the moment, but there's a strong sense of the 'me, too' about it. Anyway, head over to openxVM and see what you make of it. My strongest impression of this presentation was the contrast to the Sun keynote in 2000, given by Scott McNealy. I remember being really disappointed by Scott's home-spun style, even though some like it, and I found Jonathan a much more attractive proposition with my business hat on. I know it's about substance and not just style and I sense he's not the most popular guy, but at least I felt like he'd actually prepared for the presentation!
After that, it was off to the first of three sessions I'm attending this week coming out of the Real World Performance group at Oracle (in fact I'm typing this while I sit and wait for the last one to start).
Andrew Holdsworth's "Current Trends in Database Performance" was right up my street. It was a mild rant against the performance problems he sees at customer sites and during customer benchmarks. The more he said, the more I recognised things I tend to rant about myself. For example, most applications are implemented sub-optimally but, with the increased power of CPUs and availability of memory, people often just don't notice a problem. That's fine as far as it goes, but when you do hit a problem you're constrained by a badly written application or, worse still, a badly designed data model that eliminates any hope you have of writing an optimal application!
He talked about the difficulty of building a representative workload for capacity planning and benchmarking, which is a subject that's cropped up here in the past and I have on my list to revisit at some point.
But my favourite part was when he talked about implementing balanced systems. What I see in my travels, outside the stratosphere of Oracle's internal work and top-end customers is that sites simply don't have enough I/O bandwidth to match their other hardware resources. That cropped up again in Thursday morning's presentation too, so I'll come back to it later, but I don't doubt for a second that the Oracle Optimised Warehouse Intitiative springs from this gaping wound in customer system architecture.
Oh, no, I'm wrong. My favourite section was a discussion of how an anti-social instance can cause great pain to everyone who is using a server. A great argument for a single instance per server and I speak as someone who has just been involved with an anti-social instance! (Although in that case it's been eased by implementing AIX WLM (workload management) on it.)
The next session was the Real World Performance Group Round Table session, with Andrew (again), Mike (?), Graham Wood, Greg Rahn and the product manager. With so many bright minds on display, it was just what I was looking for and it was nice of Graham to come and say hi before they got down to business.
God, there was so much in this session it's difficult to know what to pick out, but it looks like cursor_sharing=similar should be improved in 11g according to Graham. Greg Rahn talked about how often query plan problems can be solved quickly by using dbms_xplan.display_cursor to compare estimated and actual cardinalities. Greg's obviously a really smart guy and it was a shame I hadn't put a name to the face when I met him in Chevy's the other day because I'm quite an admirer of his work so it would have been nice to meet him properly. He also highlighted the much-improved Auto-sampling code in 11g which should give very estimates in a reasonable time.
My best suggestion is that if you get a chance to hear these guys present or discuss performance with them, bite their hands off! Good stuff.
I'll write about yesterday afternoon and evening a bit later.
First was Jonathan Schwartz's keynote. You'll have read plenty about that elsewhere, but it was notable for Dell announcing Solaris support and yet another virtualisation announcement. I realise this is a hot topic for most businesses at the moment, but there's a strong sense of the 'me, too' about it. Anyway, head over to openxVM and see what you make of it. My strongest impression of this presentation was the contrast to the Sun keynote in 2000, given by Scott McNealy. I remember being really disappointed by Scott's home-spun style, even though some like it, and I found Jonathan a much more attractive proposition with my business hat on. I know it's about substance and not just style and I sense he's not the most popular guy, but at least I felt like he'd actually prepared for the presentation!
After that, it was off to the first of three sessions I'm attending this week coming out of the Real World Performance group at Oracle (in fact I'm typing this while I sit and wait for the last one to start).
Andrew Holdsworth's "Current Trends in Database Performance" was right up my street. It was a mild rant against the performance problems he sees at customer sites and during customer benchmarks. The more he said, the more I recognised things I tend to rant about myself. For example, most applications are implemented sub-optimally but, with the increased power of CPUs and availability of memory, people often just don't notice a problem. That's fine as far as it goes, but when you do hit a problem you're constrained by a badly written application or, worse still, a badly designed data model that eliminates any hope you have of writing an optimal application!
He talked about the difficulty of building a representative workload for capacity planning and benchmarking, which is a subject that's cropped up here in the past and I have on my list to revisit at some point.
But my favourite part was when he talked about implementing balanced systems. What I see in my travels, outside the stratosphere of Oracle's internal work and top-end customers is that sites simply don't have enough I/O bandwidth to match their other hardware resources. That cropped up again in Thursday morning's presentation too, so I'll come back to it later, but I don't doubt for a second that the Oracle Optimised Warehouse Intitiative springs from this gaping wound in customer system architecture.
Oh, no, I'm wrong. My favourite section was a discussion of how an anti-social instance can cause great pain to everyone who is using a server. A great argument for a single instance per server and I speak as someone who has just been involved with an anti-social instance! (Although in that case it's been eased by implementing AIX WLM (workload management) on it.)
The next session was the Real World Performance Group Round Table session, with Andrew (again), Mike (?), Graham Wood, Greg Rahn and the product manager. With so many bright minds on display, it was just what I was looking for and it was nice of Graham to come and say hi before they got down to business.
God, there was so much in this session it's difficult to know what to pick out, but it looks like cursor_sharing=similar should be improved in 11g according to Graham. Greg Rahn talked about how often query plan problems can be solved quickly by using dbms_xplan.display_cursor to compare estimated and actual cardinalities. Greg's obviously a really smart guy and it was a shame I hadn't put a name to the face when I met him in Chevy's the other day because I'm quite an admirer of his work so it would have been nice to meet him properly. He also highlighted the much-improved Auto-sampling code in 11g which should give very estimates in a reasonable time.
My best suggestion is that if you get a chance to hear these guys present or discuss performance with them, bite their hands off! Good stuff.
I'll write about yesterday afternoon and evening a bit later.
Oracle OpenWorld 2007 has come and gone and from the Real-World Performance Group’s perspective we’d consider it a successful one. The content of this year’s presentations seems to have gone over quite well as shown by the number of ...
Tracked: Nov 19, 23:59