Hotsos day Final day of the Symposium
Cary mentioned at the start of his presentation “Measure Once, Cut
Twice” that he succeeded in one of the few goals for today: Getting up
This was a nice presentation during which he showed some similarities between programming and woodwork with the aid of some nice examples. BTW it seems to me that the other presentation that he gave is available at Hotsos for registered users, and registration is free.
The main thing that I remember is that you can try to make a complete plan including all the details which makes your plan very expensive. And in the end it will most likely still has some flaws due to unforseen issues. Therefore it will be better to accept that your (project) plan cannot be perfect and instead take into account that you will need to do some ‘scribing‘ at the installation of the end product. I’m pretty sure that in time this paper will become available on the website of Hotsos as well. An other thing is that you maybe able to get needed information at a high cost early in a project that will be much cheaper to obtain a bit further down the line and with better accuracy.
The second presentation I went to was by Dr Neil Gunther, but a colleague will blog about that so I will not write about that one. Unfortunately I had to miss Robyn Sands presentation because of attending Dr Gunther’s but I hear that it was a good one. I had also wanted to attend this presentation because she was going to talk about Oracle Management and Performance from an Industrial Engineer’s view. And that appealed to me as I have an engineering background. The fact that in the handout there was an excellent paper about her presentation also made my choice a bit easier: I could always read about it later .
The third presentation I went to was by Dan Tow: “Natural Data Clustering: Why Nested Loops Win so Often”
This presentation is about why nested loops are better then has joins and when. (James Morle mentioned in his presentation that Nested Loops are classic latency multipliers, and offcourse that isn’t good). Dan talked about naturally clustered tables. That’s where recent Master rows are used together with recent Detail rows. That means that you use the recent blocks from the master table together with the recent blocks from the detail table and nested loops are a good thing apparently. But when those tables are being rebuilt using parallelism things get messed up because your recent rows end up all over the table and no longer together. (Because each of the parallel processes get a piece of the new table to create). Thus the self caching of the nested loop is gone. Idem for reversed indexes: All the recent rows end up all over the place.
He showed some experimental data to back this up.
The fourth presentation I went to was by Kyle Hailey “Average Active Sessions: a Simple Solution to Complex Performance Data”
He showed us how to drill into problems by using OEM and how much easier that was then using Tanel Poder’s scripts. But offcours you need a license to use that. It’s the same for the ASH tables, you need a license. But for his SASH tables you do not need a license nor for his OEM-lite. OEM-lite is a graphical tool that you can use to do similar things as with OEM and you can zoom into the details of the graphs by drawing a box with your mouse. Pretty cool stuff. Unfortunately he had so much to Demo and so many slides that he had to use more time then was alotted. Fortunately he was a very enthousiastic presenter and fun to watch.
The last and final presentation I saw was the one by James Morle about “Skew & Latency: The Silent Killers”
This was also a fun presentation to watch/attend and highly entertaining. Mixed with fun examples and Oracle examples the Silent Killers were explained. First a definition of latency: Latency is the wasted time associated with each operation. Most of the time latency is fixed. Bandwidth problems can be solved by upgrading your network from 1Gbit to 10Gbit but in the case of latency this is not often the case.
Most of the time you hear people say: Oh, this is such a small time/number it will not be a problem. But James had a nice way of putting it (probably not an exact quote): “Short latencies are like Carbon Monoxide molecules. They don’t look dangerous but when they accumulate then can be lethal.”
A nice live demo about latency with some beers was given between our lecture room and the one next to us. We all had a good laugh, but it was also very insightfull. It showed that bandwidth isn’t enough.
I will not go into much detail about this demo because it is much nicer to see it live for the first time then to read about it.
After that there was a wrap-up and farewell meeting hosted by Gary Goodman. For us it meant that we could let the information sink in a little while preparing for the Training Day by Tom Kyte. Yep THE Tom Kyte.
I sure hope that I can attend Hotsos next year as well, it has been an inspiring event.