Today I was working with the client loading data to MyISAM tables at very high rate. Hundreds of millions rows are loaded daily into single MySQL instance with bursts up to 100K of records/sec which need to be inserted (in the table with few indexes). It was good not all records had to go to the same table and so in theory using multiple thread to do inserts in multiple tables would not be bound by table locks and would be able to use multiple cores efficiently to get good insert rate... or so it seemed.

In reality inserting in parallel into different tables when indexes fit in memory results in very bad contention causing hundreds of thousands of context switches per second with far less work done than one would hope. As we already discussed
MyISAM key cache has serious mutex contention issue as there is global …

Last week, Mixi - Japan's largest social network service provider, launched an experimental microblogging service called Echo, and according to their delevopers' blog entry, they are using Q4M to level their write loads. Thank you to the developers of Mixi Echo for using Q4M, I hope Echo will go well and soon become a first-class service.

Meanwhile, several bugs were found in Q4M (that arise under rare cases or complex usage senarios), so I have just uploaded version 0.8, available from the Q4M homepage.

This release fixes the following bugs.

• block div-by-zero exception on conditional subcription
• do not crash when a nonexistent table …

Sun Microsystems, Inc. today announced that leading online advertising representation firm Gorilla Nation Media has selected a MySQL Enterprise™ database subscription to help power its business intelligence platform. With more than 66 million unique visitors per month, Gorilla Nation turned to Sun to help support the exponential growth of its business by improving performance and lowering costs.

[An audio interview podcast with Alex Godelman, VP of Technology for Gorilla Nation Media is available here.]

First of all, that’s to everyone who’s pitched in so far. There are more messages to translate coming… So far we have contributions in 15 different languages. Quite impressive. Check it out:

A slight engineering puzzle is coming up. As you can see from the german output, “Unknown database ‘lskfs’” wasn’t translated. This is because that error message is actually stiched together on the server and sent back over the wire, so it’s not executing in the context of the locale the client is connecting as. That’s fine for messages without parameters, but for ones with parameters, which would be the main reason to build on server and send over the wire, running gettext() on the client gets us no where. Not sure what the right answer to this is yet…

…

Ben Kaufman at SQLServerCentral introduces MySQL to the SQL Server DBA crowd. All in all his views seem to be fairly positive, in particular the MySQL Cluster experience:

NDB is the gem of MySQL, originally developed by Ericson to track cell phone calls this is a share nothing cluster engine stored in memory. This is a true cluster that supports both high availability and load balancing. [...]
This engine is similar to synchronous mirroring in SQL Server in that it is a 2-phase commit, the difference being the commit is done in memory at the data layer not the log. Logs are hardened at a later time, with the theory being that since data is committed on multiple nodes the data is safe and doesn't require a log flush as part of the transaction. [...]
For pure performance the cluster is comparable with a single instance of SQL Server. I've …

Ben Kaufman at SQLServerCentral introduces MySQL to the SQL Server DBA crowd. All in all his views seem to be fairly positive, in particular the MySQL Cluster experience:

NDB is the gem of MySQL, originally developed by Ericson to track cell phone calls this is a share nothing cluster engine stored in memory. This is a true cluster that supports both high availability and load balancing. [...]
This engine is similar to synchronous mirroring in SQL Server in that it is a 2-phase commit, the difference being the commit is done in memory at the data layer not the log. Logs are hardened at a later time, with the theory being that since data is committed on multiple nodes the data is safe and doesn't require a log flush as part of the transaction. [...]
For pure performance the cluster is comparable with a single instance of SQL Server. I've …

Dear LazyWeb, I want to use my Ubuntu laptop (on amd64 BTW) to build an RPM of Maatkit that will work on all RPM-based distros. Is it possible? Or are there enough differences between the RPM-based distros that I can’t do it? Mind you, the finished RPM ought to just have some man pages and Perl scripts, so I don’t think it will be platform- or distro-specific. But I am just not an expert on it.

The second question is, what do I need to put into my Makefile to do this? My ‘make all’ currently builds a .zip, a .tar.gz, and a .deb package — what needs to change to make that include .rpm?

Someone who is willing to help create .spec files, etc, etc will be immediately given commit rights to Maatkit’s SVN repository!

Debian, packaging, …

Introduction

The purpose of this document is to explain the power-efficeincy analysis proess, power-efficiency issues and solutions in MySQL database server as an example to help identify the right tools and procedures that support the development of energy-efficient application.

Analysis Process

1. Install and start test application(MySQL) on idle system, create workload to simulate the scenario that multiple concurrent clients connecting to the MySQL server without executing any queries.

2. Run PowerTop on Solaris to see how much time the CPUs are spending running in lower power states, and how much time the CPU are spending running in C0 state(meaning CPU is actually executing instructions.)

3. Use DTrace to profile application understand the source of the wakeups causing …

Want to know the type of an arbitrary expression in MySQL? Someday in the far far future in version 7.1, you might be able to with the TYPEOF() function.

For now you can try this:

Finally, the next release of version 5.0 is available on our download servers. During summer time, our release-cycle got a bit loose, but we’re about to tighten that again so the next release will hit the servers in less than a month again. 24 bugs were closed since release 5.0.23. Unfortunately one of the bugs discovered lately (and of course fixed in this release) was a bug in our auto-update module that could lead to a crash of Workbench when updater tried to close the running application. As a result, we had to deactivate automatic updates for versions < 5.0.24 so please fetch you copy manually this time via our download page. All customers of SE version please log into your account and download the updated SE package from there - sorry for the inconvenience.
The ongoing work on version 5.1 is making good progress, while we’re further decreasing the …