Friendster

Dathan presented about Friendster's architecture and how it evolved over time. It was the first time the talk was given. I thought the ideas were correct, but like all first talks, some of the things we focused on weren't as well-highlighted as they could have been. For example, some of the charts had too much information (the audience gave this feedback). We will likely improve the talk and give it again in a future conference.

MySQL Cluster Features and Roadmap

John David Duncan gave a talk about the cluster. This was actually one of the better talks for me, because it talked about some of the limitations of cluster now, and a roadmap for when they'd be fixed. My takeaway was 5.0 is really about the minimum version you'll want to use cluster for if you want a little more general-purpose storage engine. 5.1 will have a lot of significant performance improvements. When that version comes, cluster will likely be a very compelling storage engine.

Wednesday

Eclipse BIRT Project

Due to trying out the Views in MysQL^? 5.0 and Linux Server Configurations talks on the same day, I walked into this talk about halfway through it. It looks like a Java reporting engine that's built into the Eclipse (Java) IDE. Even if you don't use Java, it's still worth a look, because it has a lot of non-Java applications.

Building and Optimising DW Star Schemas with MySQL

Bert Scalzo talked about building datawarehouses. Nothing particularly interesting about this talk. Just another DW talk to really hammer home the concepts for me.

Tour of the MySQL Source

Monty Widenius and Brian Aker double-teamed the audience talking about MySQL internals. It was high-level enough for a non-code junkie like myself to keep up, but seemed to keep the hard-core developers entertained at the same time. There was a bit of interest in the storage engine API (API for lack of a better term) from the audience.

Multi-Terabyte Data Warehouse and MySQL

Michael Benzinger talked about Travelocity's gigantic datawarehouse. This is a huge throughput on MySQL machines. I really came away from this conference feeling confident that MySQL can do serious datawarehouse business. And since I work at Friendster, I already know it can do serious OLTP business. These types of talks show me that Oracle has a lot to fear. MySQL can scale in OLTP and OLAP. It is stable, cheap, and all the enterprise features cost the same amount as the introductory features. Tech support is better than Oracle. So... what reason is there left to choose Oracle?

Advanced InnoDB Performance Tuning

Peter Zaitsev did this talk. I do not remember Tobias Asplund being there. Peter ran through a bunch of slides talking about InnoDB performance. I believe the talk he was giving was meant for a longer 1.5 hour session, so he ran way past the end of his talk time, and even still only covered about 50% of any given slide.

Thursday

Open Source Data Warehouse

Perhaps one of the lightest DW talks at the conference. An O'Reilly guy named Roger Magoulas showed some of the tools O'Reilly uses to visualise data. It looks like the tools aren't open source, so you can't download them, but I'm unsure. I'll look for his slides to see if I can glean anything of use out of the talk.

Overview of MySQL 5.0 Query Optimiser

Timour Katchaounov talked about the current optimiser (4.x) for most of the time, if I recall correctly. Then he talked a bit about the improvements for 5.0. Actually I'm quite glad he spent more time on 4.x optimiser, since that's mainly what Friendster still uses.

Wikimedia, MySQL, and Free Software

Jimmy Wales and James Day talked about Wikipedia and sister projects like Wiktionary and Wikinews. There was a plea for help from the developer community and the admin community to help develop and admin the Wikipedia.

Tim O'Reilly's Closing Session

A lot of talk about stuff. I didn't really get a lot out of this, but I got a strong feeling that O'Reilly is all about open source and selling books, which I probably could have told you before the closing session.

The conference had two parts. Monday was broken up into two long "tutorial" sessions. You could choose from any of about 5 topics. I chose MySQL Cluster and Data Warehousing. Tuesday through Thursday were broken into many less-than-one-hour sessions. I'll detail which ones I attended below.

Monday

MySQL Cluster

Max Mether and Mikael Ronstrom led the talk. They simply went through MySQL Cluster, the terms and the setup. Toward the end of the session they set up a MySQL Cluster on their laptop and demonstrated some of the commands and usage. A fairly basic talk. The basic idea is this:

A MySQL instance is set up to use NDB storage type. There is a master node that controls things like storage node failures. There are several storage nodes that are hooked together in a ring of heartbeat. If the MySQL instance tries to store something into the NDB cluster, the storage nodes take the data and store it. If a node fails, the storage node that it reports to notices it's gone and asks the master node for an election to determine who will be primary node for the data stored on it. If it wins, it becomes primary. Storage nodes store both data they are primary for, and data they are secondary for. Secondary data on a storage node is an exact copy of the data that some other node is primary for.

Data Warehousing

John Paul Ashenfelter headed up this one. The concepts of data warehousing were covered. The idea that MyISAM is the best table type for DW in MySQL was offered up. This was echoed many times in the following days in the shorter DW talks.

The idea of a datawarehouse is to get the users behind you by offering them a service that makes their lives easier. This is a pretty obvious idea, but in the heat of implementation, apparently it often gets lost. Also, he offered that you should start with a simple datamart (ie: a single business function and not a complicated one, either) and implement it. Then, as time goes on, keep repeating the process with improvements. Each little success will give you something to learn about and more backing from the users, until you realise one day you have a full-blown datawarehouse.

Tuesday

LiveJournal^?'s Backend

The LiveJournal^? folks gave a talk about their backend. Perlbal, memcached, BigIP^?, etc. It was all entertaining and more-or-less covered the same material as their last presentation.

Managing LAMP Stacks

A guy from OSDN (OSTL?) presented some tools he uses to manage things. It was basically a long list of items in talk format. I enjoyed it but will have to find the presentation online to remember all the things he talked about. I remember that there were some things I wanted to check out, but I can't remember the names, so I'm going to punt.

Replication and Cluster Overview

Brian Aker talked about MySQL Cluster again. Basically this was the talk for those that didn't make Monday's session, so it covered a lot of the same material.

Panel: Scaling and High Availability Challenges

This was a session where a bunch of industry mavens congratulated themselves in a rambling, confused way.