[en] on SWiK

A Wordpress MU Based Social Network Platform | buddypress.org

Sat, 30 Aug 2008 05:57:29 -0700

buddypress is a wordpress a social platform you can download and based on wordpress MU

The Git and the Bazaar | kosara.net

Mon, 18 Aug 2008 07:06:07 -0700

While the former [Git] may be technologically the most elegant and efficient system around, Bazaar is so far ahead in usability, documentation, and support that the choice is really quite easy.

flyspell.el

Sun, 10 Aug 2008 11:07:44 -0700

... starting of Flyspell when editing LaTeX source files, one may add the following expression in his .emacs file: (add-hook 'LaTeX-mode-hook 'flyspell-mode)

Emacs for vi users

Sat, 09 Aug 2008 12:08:14 -0700

... for vi users wishing to learn the basics of Emacs. I'm writing it because I'm one of them... will not list features that are specific to Emacs. I believe there to be sufficient overlap between...

linux:devilspie [foosel.(net|org)]

Tue, 15 Jul 2008 05:50:32 -0700

Cleanly Migrate Your Subversion Repository To a GIT Repository | simplisticcomplexity.com

Wed, 25 Jun 2008 13:55:42 -0700

Nice tips on how to migrate from svn to git repository

A Preview of HTML 5 | alistapart.com

Mon, 23 Jun 2008 15:01:54 -0700

A very complete article explaining what there will be in HTML5

Planet HTML5 | w3.org

Mon, 23 Jun 2008 15:01:53 -0700

The planet agregating many sources about HTML5

Instructions for joining the HTML Working Group | w3.org

Mon, 23 Jun 2008 09:02:11 -0700

Instructions for joining the HTML Working Group

An HTML 5 presentation | james.html5.org | [PDF]

Mon, 23 Jun 2008 10:01:33 -0700

An HTML5 presentation done by James Graham at a barcamp

The Importance of Being HTML5 | snook.ca

Mon, 23 Jun 2008 10:01:33 -0700

An article about HMTL5 backward compatibility and lots of comments

Six Months Later: The New HTML Working Group | sitepoint.com

Mon, 23 Jun 2008 10:01:32 -0700

An article written just after the creation of the new HTML working group and the acceptance of the WHATWG HTML specs

HTML5 Google suggest | simon.html5.org

Mon, 23 Jun 2008 10:01:31 -0700

Another test for a google suggest with HTML5's specs

Big Buck Bunny

Mon, 02 Jun 2008 16:17:45 -0700

Court métrage en images de synthèse fait entièrement à l'aide de logiciels libres.

doctype - doctype - Google Code

Thu, 15 May 2008 10:21:28 -0700

Doctype is a Google-sponsored open encyclopedia and reference library for developers of web applications. By web developers, for web developers.

Testing InnoDB and MySQL 5.1 performance in real workload conditions

Fri, 02 May 2008 12:12:44 -0700

Since Heikki Tuuri and Peter Zaitsev’s presentations at the MySQL User Conference 2008, I really wanted to try MySQL 5.1 with InnoDB plugin in production work to see how it compares to my current setups (MySQL 5.0.48 with integrated InnoDB).

First of all, the upgrade to MySQL 5.1.23-rc went without any particular quirks - I did not have to dump and reload tables, a simple mysql_upgrade script run was sufficient to put my DB to MySQL 5.1-readiness.

When starting a new MySQL instance you know that you have to do some warm-up work before getting any indicators - the buffer pool will be empty of indexes and data so most operations in the first hour of uptime will be IO-bound. You can also use clever queries to speed warm-up time as indicated by Peter Zaitsev on his blog.

Workload Characteristics

It’s always good to know that benchmarks and system or software upgrades can give different results depending of whether you are I/O-bound or CPU-bound and of your hardware platform characteristics.

In our particular test case, the workload is mostly I/O bound (available InnoDB buffer size is about 1/8 of exploitable dataset size) and the hardware platform is composed of 2 x dual-Core Xeon 5148 with 16G of DDR, and 8×15k SAS drives in RAID-0 mode (BBU controller, write-back enabled).

Average thread concurrency is always around the 32 threads mark which is good for our benchmark case because we want to see how MySQL 5.1 and InnoDB perform in superior workload conditions. The queries’ workload is mostly read-oriented : we’re in a case of a traditional web 2.0 database with very few writes and lots of reads.

Production test

The first indicator that really interests me in this case is how the database performs with current workload - QPS (Query per Second) is generally a good indicator and is available in many tools like innotop.

MySQL 5.0.40 + built-in InnoDB: 2.0k QPS avg.
MySQL 5.1.23-rc + InnoDB plugin 1.0: 1.5k QPS avg.

As you can see there is a serious performance drop with MySQL 5.1 (about 25% less QPS) which corroborates Peter Zaitsev’s slides from the MySQL UC, noting some potential regressions in 5.1 which lead to a noticeable decrease of performance.

Another indicator which was disappointing was the number of SELECT queries per second, which we graph using RRDTool.

For MySQL 5.0 the number of SELECT per second is usually steady around the 1.2k mark.

With this new installation of MySQL 5.1 and InnoDB, this number was no higher than 900k SELECTs, and the database showed signs of thread congestion at some times : performance dropped to half this number during a few minutes before attaining peak again.

Table compression test

After this rather disappointing start, I wanted to try a new feature of InnoDB plugin : table compression. I went for a 8k compressed page size according to Venu Anuganti’s tests and compressed a 6GB table which is one of the most used tables in our application. The table compression rate was interesting at 73% (I used the method recommended by Ken Jacobs at the MySQL UC: gzip a n.ibd file and see if the resulting gzip file is more than 50% compressed)

In production, the results were promising in CPU usage since we switched from an I/O-bound workload to a totally CPU-bound workload.

Before (no compression) : 30% user CPU, 30% iowait
After (using 8k compression on the main table) : 87% user CPU, 3% iowait

Impressive numbers! But there was an outstanding problem with compression: some requests greatly underperformed.

Actually, we have in our application, some requests which I will call “external sorts” which are used to build various rankings. i.e. we do this kind of joins:

mysql> SELECT id FROM main INNER JOIN category USING (id) INNER JOIN ranking ORDER BY ranking.total LIMIT XX; *************************** 1. row *************************** id: 1 select_type: SIMPLE table: category type: index possible_keys: PRIMARY key: PRIMARY key_len: 4 ref: NULL rows: 750465 Extra: Using index; Using temporary; Using filesort *************************** 2. row *************************** id: 1 select_type: SIMPLE table: ranking type: eq_ref possible_keys: id key: id key_len: 4 ref: id rows: 1 Extra: *************************** 3. row *************************** id: 1 select_type: SIMPLE table: main type: eq_ref possible_keys: PRIMARY key: PRIMARY key_len: 4 ref: id rows: 1 Extra: Using where; Using index

The optimizer will usually reverse table execution order because of the ORDER BY clause and will fetch all of the ids in the category table, then do a temporary and filesort to build the final resultset. This will usually work fine with no compression (a few secs) and will be OK on the application side provided that you store the resultset in memcached.

With compression, I guess that the performance problem is derived from the fact that MySQL will not reading adjacent pages of data in the main table, but that it will be forced to fetch and decompress many non-adjacent pages due to the fact that the main table is not sorted by index, but rather by filesort algorithm. I know that filesort + temporary is not very good query design on large pages of data but for us it was a tradeoff which worked without using compression - looks like I’ll have to make more benchmarking tests on filesorts + compressed data to assess if the decrease of performance is at this particular spot.

Jason Blevins: Emacs markdown-mode

Thu, 21 Feb 2008 03:43:25 -0800

HTML 5 differences from HTML 4

Thu, 24 Jan 2008 12:07:56 -0800

Tamper Data | addons.mozilla.org

Tue, 22 Jan 2008 16:27:41 -0800

A firefox plugin to snif and filter every request (post/get) made by the browser. Also give you a nice tool to modify all headers and parameters. Even lets you add and remove them and re-run the request

A Short History of BPM, Part 1 - Column 2 - ebizQ

Thu, 17 Jan 2008 04:52:52 -0800

Geschichte von Business Process Management in 8 Teilen | 2006 - 2007

A Short History of BPM, Part 1 - Column 2 - ebizQ

Thu, 17 Jan 2008 04:40:23 -0800

Geschichte von Business Process Management in 8 Teilen | 2006 - 2007

Firebug

Wed, 09 Jan 2008 14:30:53 -0800

Calendar: a Javascript class for Mootools that adds accessible and unobtrusive date pickers to your form elements

Wed, 09 Jan 2008 12:42:33 -0800

Une librairie JS pour Mootools pour afficher un calendrier

HandBrake

Fri, 14 Dec 2007 02:03:14 -0800

Convertisseur de DVD vers iPod open source, multiplateformes.

Scandalous Software | scan.dalo.us

Thu, 13 Dec 2007 13:08:52 -0800

A nice lice of free mac softwares to use with XML, XSL, SOAP, XML-RPC, XQuery, XSLT

FireGPG - Install the extension

Wed, 12 Dec 2007 14:39:49 -0800

PWGen Project Homepage

Thu, 06 Dec 2007 02:33:44 -0800

Un générateur de mot de passe

W3C HTML

Tue, 04 Dec 2007 11:39:00 -0800

HTML 4.01 Specification

Tue, 04 Dec 2007 08:39:01 -0800

W3C Feed Validation Service, for Atom and RSS

Tue, 04 Dec 2007 08:39:00 -0800