The SnapLogic.org site runs on open source software, including a lot of Python. The bottom of the stack is RedHat Linux, the web server is Apache, the database engine is MySQL, and the main applications are Mailman, Subversion, Trac, Wordpress, and a custom Django application.

All of these pieces play together pretty well, and over the past year we’ve only had a few minor problems to fix. But over the past couple of weeks, we’ve really been hitting Trac hard as we work through our upcoming release, and Trac performance was bad. Really Bad. Like, ‘30 seconds to open a ticket‘ bad.

It’s hard to say if this was a sudden performance drop because we crossed some magic threshold in database size, or if it was slowing getting worse and we really noticed it when we shifted from design to coding. In any case, there are only 1000 tickets in the system so we knew something was wrong.

We turned on the …

Maximizing Sysbench OLTP performance for MySQL

Sysbench is a popular open source benchmark used to measure performance of various parts of the operating systems and (one) application (database). Since we are talking about MySQL, I will concentrate on the database part of the test.

The oltp test in the sysbench benchmark creates a single table and runs a set of queries against it. Each row in the table is around 250 bytes and by default it creates 1000 rows. For our experiment we used 10 million rows. Allan has blogged about the details regarding the experiments; I will present an alternate view to those experiments.

The trick to getting good numbers with Sysbench and MySQL is very simple

1. Maximize CPU utilization
2. Reduce delays due to IO
…

Maximizing Sysbench OLTP performance for MySQL

Sysbench is a popular open source benchmark used to measure performance of various parts of the operating systems and (one) application (database). Since we are talking about MySQL, I will concentrate on the database part of the test.

The oltp test in the sysbench benchmark creates a single table and runs a set of queries against it. Each row in the table is around 250 bytes and by default it creates 1000 rows. For our experiment we used 10 million rows. Allan has blogged about the details regarding the experiments; I will present an alternate view to those experiments.

The trick to getting good numbers with Sysbench and MySQL is very simple

1. Maximize CPU utilization
2. Reduce delays due to IO
…

Maximizing Sysbench OLTP performance for MySQL

Sysbench is a popular open source benchmark used to measure performance of various parts of the operating systems and (one) application (database). Since we are talking about MySQL, I will concentrate on the database part of the test.

The oltp test in the sysbench benchmark creates a single table and runs a set of queries against it. Each row in the table is around 250 bytes and by default it creates 1000 rows. For our experiment we used 10 million rows. Allan has blogged about the details regarding the experiments; I will present an alternate view to those experiments.

The trick to getting good numbers with Sysbench and MySQL is very simple

1. Maximize CPU utilization
2. Reduce delays due to IO
…

Recently we put together a consolidation benchmark to see how an open-source stack performs against the proprietary stack from Microsoft. Solaris, MySQL, and Sun Web Server running on the open-source UltraSPARC T2 processor were pitted against a Microsoft SW stack running on a 4-socket QC Xeon server. This benchmark highlights the continued trend to incorporate MySQL open-source databases and how it works under virtualization (Solaris Zones).

The Sun SPARC Enterprise T5220 (1.4 Ghz UltraSPARC T2 processor) and Solaris Containers managing a consolidation of Open-Source Software components (MySQL Database and Sun Java System Web Server) provided 2.4 times better performance than the HP DL580 system (four Xeon quad-core processors) and a major virtualization software, Microsoft Windows 2003 Server EE, Microsoft SQLserver database and Microsoft IIS webserver.

The Sun SPARC Enterprise T5220 using the MySQL database in Solaris zones is …

Recently we put together a consolidation benchmark to see how an open-source stack performs against the proprietary stack from Microsoft. Solaris, MySQL, and Sun Web Server running on the open-source UltraSPARC T2 processor were pitted against a Microsoft SW stack running on a 4-socket QC Xeon server. This benchmark highlights the continued trend to incorporate MySQL open-source databases and how it works under virtualization (Solaris Zones).

The Sun SPARC Enterprise T5220 (1.4 Ghz UltraSPARC T2 processor) and Solaris Containers managing a consolidation of Open-Source Software components (MySQL Database and Sun Java System Web Server) provided 2.4 times better performance than the HP DL580 system (four Xeon quad-core processors) and a major virtualization software, Microsoft Windows 2003 Server EE, Microsoft SQLserver database and Microsoft IIS webserver.

The Sun SPARC Enterprise T5220 using the MySQL database in Solaris zones is …

Recently we put together a consolidation benchmark to see how an open-source stack performs against the proprietary stack from Microsoft. Solaris, MySQL, and Sun Web Server running on the open-source UltraSPARC T2 processor were pitted against a Microsoft SW stack running on a 4-socket QC Xeon server. This benchmark highlights the continued trend to incorporate MySQL open-source databases and how it works under virtualization (Solaris Zones).

The Sun SPARC Enterprise T5220 (1.4 Ghz UltraSPARC T2 processor) and Solaris Containers managing a consolidation of Open-Source Software components (MySQL Database and Sun Java System Web Server) provided 2.4 times better performance than the HP DL580 system (four Xeon quad-core processors) and a major virtualization software, Microsoft Windows 2003 Server EE, Microsoft SQLserver database and Microsoft IIS webserver.

The Sun SPARC Enterprise T5220 using the MySQL database in Solaris zones is …

Some people might have noticed that Workbench runs very slowly and flickers a lot in their machines. Other people will be able to work without noticing any significant sluggishness with the diagram graphics. The difference is because Workbench’s custom canvas may use hardware accelerated OpenGL graphics or non-accelerated software rendering, depending on the system (ie, whether or not a 3D accelerated graphics card is available or not). We were expecting that the most reasonably recent machines (maybe 5 years old or less) would have at least some basic graphics acceleration, but for some reason, that seems to not always be the case. Apparently, some people that have decent systems with ATI or Intel cards seem to get the fallback software rendered canvas, which will result in sluggish graphics.

We are still not sure why Cairo/Glitz (the underlying graphics library used by our drawing …

How often do we think about our http sessions implementation? I mean, do you know, how your currently used sessions-related code will behave when sessions number in your database will grow up to millions (or, even, hundreds of millions) of records? This is one of the things we do not think about. But if you’ll think about it, you’ll notice, that 99% of your session-related operations are read-only and 99% of your sessions writes are not needed. Almost all your sessions table records have the same information: session_id and serialized empty session in the data field.

Looking at this sessions-related situation we have created really simple (and, at the same time, really useful for large Rails projects) plugin, which replaces ActiveRecord-based session store and makes sessions much more effective. Below you can find some information about implementation details and decisions we’ve made in this plugin, but if you just want to try it, then …

I'm pleased to announce that Tag1 Consulting has partnered up with MySQL AB to offer an online presentation titled "Achieving Optimal MySQL Performance For Drupal". Aiming to provide a better understanding of how to properly monitor and tune your MySQL database, the online Webinar will take place on Thursday, January 31st, 2008, at 16:00 UTC (11:00 am EST). The presentation will last 45 minutes, followed by 15 minutes for questions and answers.

read more