A Day of Real World Performance

Tuesday was a fun day this week as Graham Wood, Tom Kyte and Andrew Holdsworth visited London for the latest Real World Performance Day.

This is an event that has already run several times in the US via the IOUG and consists of a three-person presentation style which I thought worked very well - much better than I expected it to if I'm honest. Each presenter played a different role and often there were slightly different perspectives. Truth be told, I have quite a low boredom threshold sometimes and even with the long presentation slots I didn't find myself drifting off. Multiple top-notch speakers definitely helped.

Regular readers of this blog will be very familiar with most of these chaps. Perhaps less so Andrew Holdsworth who heads up the Real World Performance group at Oracle. If you attend Oracle Openworld you'll be familiar with him but I'm not sure I've ever seen him present in the UK. A lot of the slides would have been familiar to OOW Real World Performance presentation attendees too but, although I might not have learned too much, I preferred the way it was presented here. The day was split up into Data Warehouse-focussed material in the morning and OLTP in the afternoon, which was a useful way of covering a lot of topics whilst highlighting different performance goals.

As most of the people from my current site who attended are predominantly focussed on DW, the morning probably proved more useful. The only slight criticism I might make is that, despite the fact that Andrew was very clear that the first step that everything else was predicated on should be a balanced hardware configuration with enough power, that was easy to forget as the morning went on. So I heard a few people bemoaning the fact that all of the demos were on Exadata and so they wanted to know how much of the stunning improvements were down to Exadata and how much to the approaches used. Personally, I didn't think the absolute numbers mattered so much as the principles illustrated and I thought it was clear that enough power in a balanced configuration was the foundation, whether that's Exadata or (heaven help us!) another platform

My feeling was that this might not be everyone's Real World Performance on the systems that they're working on which is what perhaps confused people, but was the performance that *should* be possible in the real world if you buy enough kit and use the correct approach. If you've worked on badly configured systems, your own experience might be very different to what's achievable these days. To give you an example, it probably is the case these days that your data loads should be CPU-bound and that compression should be what's limiting your throughput, but that's predicated on not having a useless storage subsystem. I can certainly recall situations where loads were I/O bound so compression was used to speed the loads up and they were *still* I/O bound. By always keeping in mind 'this is what we should be aiming for', I thought it was all very useful and achievable. Perhaps worth reminding people a little more often that some of the techniques might not deliver if your configuration is wrong but, I repeat, Andrew was very clear that this is the *first* step.

Sadly if you insist that all of your data has to be fed in through Java application servers that pretty much undermines a lot of the morning content, around the Terabyte data loading exercise, but all you can do is try to lead people in the right direction!

One aspect of the Real World Performance presentations I've always appreciated is that they discuss ball-park expectations you should have of different system components and configurations. I know from personal experience of teaching performance courses that many people lack this information and are hungry for it. Fortunately the material is publicly available in a zip file on Tom's site so I don't need to cover the technical details. (Note, however, that this is just the slide material and doesn't cover the demos, the questions and answers and dialogue so it's a poor substitute for the experience of attending.)

There's much more I could say about the day. I know a couple of colleagues who enjoyed it as much as I did and found it useful (some more than others) to the extent that I think they might be attending a few more conferences in future if they're this good

Congratulations to all involved and I'll finish with a piece of good news for those who missed it. According to Thomas Presslie (@tpresslie) there will be another run of the same event in Edinburgh on the 23rd August 2011. Not to be missed!