FOSDEM 2011 Developer Room
There will be a MySQL Developer Room at FOSDEM 2011 on Saturday, 5th of February, providing MySQL-related talks and sessions.
The room number is H.2213 and has a capacity of 100 seats (in building "H"). It is equipped with a video projector with VGA cable and Internet connectivity (wireless only).
[edit] Recordings / Transcripts
For people that could not make it to FOSDEM or any of the sessions, it would be nice to be able to provide video recordings or transcriptions (e.g. live blogging) from these talks. We're looking for volunteers to record the sessions!
[edit] Session schedule
This page will provide the session schedule for the "MySQL and Friends" Developer Room at FOSDEM 2011. Each talk will last 20 minutes, plus 5 minutes for Q&A. There will be a 5 minute break between each session, to allow people to leave/enter the room and the speakers to get hooked up. The final schedule will available from the official FOSDEM schedule as well. However, this page is the authoritative source (as the FOSDEM version may not always be up to date with the latest changes).
PDFs or links to the presentation slides will later be added to this page. Click on the speaker name for the speaker's biography. Clicking on the session title will get you to the session abstract.
[edit] Abstracts
[edit] Istvan Podor: Data Recovery for MySQL
MySQL is a modern stable database, however data can be lost for a number of reasons. This session discusses the different types of data loss and recovery techniques and how to use the Percona InnoDB Recovery Tool to recover your data.
[edit] Kenny Gryp: Taking Backups With XtraBackup
XtraBackup is an open-source online backup solution for InnoDB/XtraDB databases with support for MyISAM.
The session will cover all the current features XtraBackup has:
- compressed backups
- use filters to choose which tables/schemas to backup
- stream backups to another server
- export/importing of tables
- incremental backups
- throttling backups to a certain amount of IO requests
[edit] Steve Hardy: Boosting Enterprise MySQL performance: implementing I/O prefetch for InnoDB
Tests have shown that InnoDB is not using traditional spindle-based I/O subsystems to their maximum potential while doing OLTP workloads with I/O-bound queries; Pages are requested sequentially from the disk system instead of asynchronously, even when using MySQL 5.5's native Aynchronous I/O support. The presentation shows the advantages of implementing a prefetch mechanism which can result in large gains in I/O performance.
For more information of the Innodb prefetch improvement see this mailinglist post: http://lists.mysql.com/internals/38126
[edit] Petya Kohts: Multimaster MySQL failover
I propose the algorithm and proof of concept tool which solve one of key aspects of scaling mysql read load with multiple read slaves: failover in case of master failure.
Algorithm works in common configuration with master serving writes and one or more slaves serving reads and replicating from master.
The algorithm allows any of read slaves to become new master and guarantees full consistency of all the slaves which were available during its work.
In short: algorithm allows the same degree of consistency as a popular dual master (master-master) mysql replication scheme adding the ability to use a number of slaves for read load and failover as compared to one in the dual master replication scheme.
[edit] Lenz Grimmer: MySQL 5.5 Replication Enhancements: an overview
There have been quite a number of improvements and enhancements to the replication functionality in MySQL 5.5. Starting from the big ones like semisynchronous replication and the replication heartbeat, this talk is taking a closer look at the changes in this area.
[edit] Lenz Grimmer: What's new in MySQL 5.5
In this talk, I would like to give a "helicopter view" of the new features and enhancements that have been added to MySQL 5.5. The talk will broadly cover all notable changes that have been made in 5.5 in comparison to previous releases.
[edit] Istvan Podor: Over 20,000QPS, XtraDB performance show
In this short 20 minute talk I would like to give you a quick look into the process I've optimized read-slaves for a high traffic website (Alexa top500) using xtradb. When I started to work on this project the slaves were able to serve like 5-8k QPS with a basically default configuration. When I was done, in peak times we goes over 20,000QPS having a low load. I would like to use my 20 minutes to show how I optimized the mysql server and show what effects each optimization had and how we tuned a standard mysql replica slave to a beast. In this presentation I don't want to mention query tuning.
[edit] Giuseppe Maxia/Linas Virbalas: Seamless MySQL replication failover with Tungsten
Replication fail over and fail back are the most common occurrences in MySQL database operations. And yet, they are still among the most difficult tasks to perform, all because MySQL lacks the ability of matching master transactions with the ones stored in the slaves.
Since Tungsten has the global transaction ID, promoting a slave to master and synchronizing the other slaves is a painless operation.
This session will show the difference between switching between master and slave in two scenarios: using MySQL regular replication, and using Tungsten replication.
[edit] Stéphane Combaudon: A practical overview of Maatkit
"Maatkit has saved my day more than once. In this talk, I will show you real-life examples of hard problems that I had to solve, and how Maatkit provided elegant (and most of the time very efficient) solutions.
I will cover the following short stories:
- When I run this query manually, I get an immediate answer, but when the same query is run by the application, it takes 30+s. Why?
- Did you know that the data you're backing up is not the right data?
- You say that we're running 400 SET queries/s. I doubt this, could we catch them live?
- Last time I wanted to update this field in the whole table, it took 2 weeks. Any tips to run the update faster?
- I had to skip an event on a slave. To avoid inconsistencies with the master, is rebuilding this 10GB table from scratch the only way?
- We regularly have very long queries that freeze the whole server. Could we kill them automatically if they run for more than 50s?
And a few more if time permits.
[edit] Oli Sennhauser: HandlerSocket and similar technologies - NoSQL for MySQL
Yoshinori Matsunobu’s blog post hit the MySQL World like a bomb! He claimed that he got 750k PK lookups per second ""on a commodity MySQL server"" with the InnoDB Storage Engine this comes pretty close to the 1M Transactions per second the MySQL Cluster guys claimed for a 10-node Cluster earlier in 2009/2010.
Other Projects have implemented such methods already 3 years ago with the BLOB Streaming Engine from Primebase.
In this presentation we have a closer look at how those technologies work and how to set-up such a beast to get similar numbers. Where we can/should use it and where not.
[edit] Seppo Jaakola: Multi-Master Replication Approaches
"Aside from traditional master-slave setup used mostly for HA and read scale-out, there is an established history of attempts at multi-master replication with MySQL. In the presentation, we discuss the use cases for multi-master replication:
- write scaling
- simplified load balancing
- latency reduction by moving server closer to clients (a case for WAN multi-master)
The presentation will compare how various multi-master replication approaches like MMM, DRDB, NDB and Galera fare with regards to scalability, high-availability and transparency for application. How these solutions detect conflicts and resolve them. And how much application should be involved in resolving or avoiding multi-master conflicts.
[edit] Sergey Petrunya: New query optimizer features in MariaDB
Query optimizer is one of the focus areas in MariaDB development. This talk will give an overview of the recently completed (but not yet presented anywhere) optimizer features:
- index_merge intersection
- Key ordered retrieval for disk-sweep MRR
- Batched Key Access improvements
- [Classic] Hash join
for each feature, I'll provide an easy-to-understand explanation of what it is and how we expect it to be useful. Then I'll show a mini-benchmark, and discuss current limitations and possible gotchas.
[edit] Speaker Information
This paragraph will provide some background information about the DevRoom speakers.
[edit] Stéphane Combaudon
Stéphane is a MySQL DBA at Dailymotion, one of the leading video-sharing websites in the world. He has been using the LAMP platform for almost 10 years in various companies and is particularly enthusiastic about MySQL. He is a regular speaker at conferences on MySQL and on free software and is a member of the MySQL User Group France.
[edit] Lenz Grimmer
Lenz Grimmer is a member of the MySQL Community Relations Team at Oracle. He lives in Hamburg, Germany and has worked for MySQL since April, 2002. Before joining the Community Relations team in December 2005, he was a member of the Release Engineering team that is in charge of creating the official release builds of the MySQL server. Prior to working for MySQL, Lenz was an engineer in the distribution development team of SuSE Linux in Nuremberg for four years. He has a degree in Computer Science and has been involved in Linux and Open Source since 1994.
[edit] Kenny Gryp
Kenny joined Percona in July 2010. He came from online social network NetLog, where he was the MySQL DBA. He started using open source long before the millenium and has been involved in system administration for more than 7 years.
[edit] Steve Hardy
Steve Hardy is CTO of Zarafa, an open-source groupware vendor using MySQL as storage platform. Steve has been working on MySQL optimization since Zarafa's introduction in 2005, both by improving schemas and queries and by developing inside the MySQL codebase.
[edit] Seppo Jaakola
CEO of Codership, long experience in developing DBMS replication and clustering solutions.
Seppo works as CEO at Codership, focusing in development of high performance DBMS replication, clustering and high availability solutions. His main responsibility is in integrating Galera replication service with DBMS engines.
Seppo has a long experience in distributed computing, clustering solutions and DBMS development. His earlier employments include ~10 year assigment in Stonesoft R&D developing firewall clustering and main memory DBMS products and ~4 year assignment in Continuent R&D, implementing clustering aware version of MySQL server.
Seppo has MSc degree in Software Engineering from Helsinki University of Technology.
[edit] Petya Kohts
An independent Internet professional with a career spanning from leadership in a small business IT department to being a key member of a management software development team in Yandex, the leading russian search engine. Contributed to nagios, zabbix and perl core modules. Author of snaked, cron replacement.
[edit] Giuseppe Maxia
Giuseppe Maxia, a.k.a. “The Data Charmer” works as QA Director at Continuent. He’s an active member of the MySQL community and long timer open source enthusiast.
During the past 23 years, he has worked in various IT related fields, with focus on databases, object oriented programming, system administration.
Fluent in Italian, English, Perl, Python, SQL, Lua, C, Bash, and good speaker of C++, French, Spanish, Java. Works in cyberspace, with a virtual team.
[edit] Sergey Petrunya
Sergey Petrunya is a developer at Monty Program Ab, working on query optimizer improvements in MariaDB. He has coded Table Elimination and improved Disk-Sweep Multi Range Read features, and also was involved in work on other optimizations.
Prior to working at MariaDB, he has been a member of query optimizer team at Sun/MySQL and has worked on index_merge access, partition pruning, and many other smaller optimizer features.
[edit] Istvan Podor
Having worked with several Alexa top500 websites from all around the world, Istvan is now part of the consulting team at Percona. With extensive experience in LAMP engineering, HA solutions, MySQL optimization and performance tuning in general behind high traffic websites. This talk will go through one of his favorite and most passionately made pastimes, tuning a MySQL backend to serve more like it should be able to.
[edit] Oli Sennhauser
Oli is a Senior MySQL Consultant and the founder of FromDual. He works with databases since the 2nd half of the Ninety's. Most of the time as consultant. Since 2006 he works as MySQL consultant for many customers mostly in Europe. He likes solving customers problems in various areas like MySQL operations, MySQL Cluster, Performance Tuning, Architecture and Design etc.
[edit] Abstracts of rejected proposals
The following talk proposals have been rejected by the reviewers. They are being kept here for reference purposes.
[edit] Roland Bouman: MQL-to-SQL: a JSON-based query language for RDBMS access from AJAX applications
MQL-to-SQL is an Open Source project allowing you to query a RDBMS over HTTP using MQL (Metaweb Query Language). MQL was designed for www.freebase.com, but there are very good reasons to want something like MQL as a RDBMS query language as well. This talk explains what MQL is, how it compares to SQL, why it's great for AJAX apps, and how MQL-to-SQL allows you to query popular RDBMS-es (like MySQL)
Past year, I got interested in Freebase and its query language, MQL. In my opinion, MQL is an interesting database query language in general, and a number of features make it downright excellent for implementing secure but flexible access to relational databases over the web, especially for modern (AJAX) web applications - better than SQL.
I created an implementation dubbed MQL-to-SQL, which is available at http://code.google.com/p/mql-to-sql/ under the terms of the LGPL. It's basically a PHP script that supports the the protocol of the freebase MQL-read webservice. It abstracts data access using PDO, and I have working examples for the sakila sample database working on SQlite, MySQL, Postgres (pagila) and Oracle. For sample queries, see the project wiki (http://code.google.com/p/mql-to-sql/wiki/SampleMQLQueries). To kickstart it without downloading and installing, try them in the online demo (http://mql.qbmetrix.com/mqlread/mql-to-sql-query-editor.php)
In this I will explain how MQL works and why it is a great solution to solve the RDBMS access problem for modern (AJAX) web applications. In addition, I will explain how I implemented MQL-to-SQL, and how it translates MQL queries to SQL queries, executes them and returns the result document. Time permitting I will show what the advantage is for the web-client (as compared to being limited to tabular results returned by SQL queries)
Solving the data access problem in Web vs client-server applications:
- How web applications typically use a purpose-built REST or RPC style webservice and a data format like XML or JSON to access and manipulate data over HTTP
- How by contrast, client-server applications typically use a query language like SQL instead of a purpose-built service protocol
- How this relates to so-called document-based queries and results in NoSQL solutions like CouchDB and MongoDB
The problem with SQL from the perspective of a web applications
- Why you don't very often see a webservice that accepts SQL queries
- How to integrate SQL with your application language, and how ORMs make it even worse (yes!)
The MQL query language
- Some background on freebase, for which MQL was designed
- Syntax of MQL queries and results, with examples (hello world and beyond)
- SQL vs MQL: what is similar, what is not; what are the respective strengths and weaknesses and how does that translate to particular use-cases
Why MQL is a great solution for modern (AJAX) web applications:
- accessible through a webservice, but because it is a query language, it is inherently flexible, and you don't need a purpose-built webservice.
- Application language integration: both queries and results are JSON-based, which is easily processed and generated by JavaScript, which is the de-facto client side and emerging server-side scripting language for web applications
- Query by example and Query/Result symmetry: in MQL, generating new queries from previously returned results is trivial. This is very unlike the SQL case, where the language is an algebra, but the result is pure data.
- without compromising flexibility, MQL is much easier to secure than SQL
MQL-to-SQL: an open source MQL implementation for your RDMS
- concepts: mapping MQL to SQL and relational results to result documents
- implementation: PHP, PDO
- how to get started with MQL-to-SQL using your favorite RDBMS
[edit] Lukas Smith: MyISAM "rsync-based replication"
During development of a project we were faced with a 5GB+ CSV data dump and nightly delta updates varying greatly in size. More importantly the CSV data has a fairly high risk of containing serious data consistency issues. Finally the data is read only. So in the end we decided to use MyISAM tables, run the import on one server overnight and then ship the full MyISAM tables via rsync to the other slaves. This way we can ensure that only if the import has run through cleanly we apply it to the slaves. It works, though right now are only in "silent launch" mode. This talk will present the solution and I assume will lead to a lot of critical questions and potentially the speaker accepting that a subpar solution was implemented.
[edit] Oli Sennhauser: Virtualization the MySQL way - get rid of the overhead!
Virtualization is everywhere at the moment. Virtualization makes sense when you want to consolidate your resources. Basically it means: Less hardware for the same amount of service.
The typically approach is to use some kind of virtualization solutions like VMWare, Virtual Box, Xen etc. If you want to consolidate just your MySQL databases this is a complete overkill and can be achieved much easier by multi-instance set-up's.
In this talk we look at the pros and cons of the multi-instance set-up method, where we can/should use it and where not.
[edit] Sergey Petrunya: Differences between MariaDB and MySQL releases
Development around MySQL is no longer centralized. Multiple parties are producing new features and patches, as well as bundling them into releases. What does one get (or miss) when they install a particular version of MariaDB or MySQL?
This talk will give a feature-wise comparison of stable MySQL and MariaDB releases that will be easier to comprehend than diffs of release changelogs. A list of pointers to published benchmarks between versions will also be given.