Bull’s Implementation of a Glued Architecture


In my two previous posts, I’ve been introducing the concept of ‘glueless’ and ‘glued’ as the two main scale-up architectures. You can read them here and here. Eventually you may also read this post in the series talking about the need to go now for a scale-up approach to virtualize the last bit, that is resource-hungry business and mission critical applications.

We’ve seen that the ‘glued’ architecture is the best architecture choice to scale-up beyond 4- and 8-socket systems. We’ve also noticed that the quality of the OEM-developed eXternal Node-Controllers is critical.

Meet the Bull Coherence Switch Architecture. BCS Architecture is Bull’s implementation of the glued eXternal Node-Controller. It is the design foundation for Bullion x86 servers that need to deliver more scalability, resiliency, and efficiency to meet requirements of the most demanding applications in the business computing.

Bullion's BCS

16-socket glued architecture using Bull’s BCS technology– Courtesy of Bull

A bit of history. The BCS technology is the foundation of bullx Supernodes series of supercomputers designed to run HPC applications that require huge volume of shared resources, in particular shared memory.

Bull decided to leverage that technology into their bullion series pushing the limit of x86 enterprise-class servers to a new level.

In July 2012, the bullion server has been ranked as the world’s fastest x86 enterprise-class server, according to the international SPECint®_rate2006 benchmark.

Featuring 160 Intel® Xeon® E7 cores and 4 Terabytes of RAM, the bullion server achieved peak performance of 4,110 according to the SPECint®2006 benchmark. The fastest competitive system, HP Proliant DL980 G7, only managed a performance of 2,180.

bullion-Intel-x86

The bullion server achieved peak performance of 4,110 according to the SPECint®2006 benchmark – Courtesy of Bull

Bullion SPECint

SPECint®_rate2006 data, July 2012

These results show that not only Bull’s ‘glued’ architecture is the way to go for scale-up architecture, but also that Bull engineered a master piece of technology, that is the BCS.

Remember that one the main drawbacks of the ‘glueless’ architecture is that up to 65% of Intel QPI links bandwidth is consumed to address QPI source broadcast snoopy protocol, that is  maintaining cache coherency when socket increases. The performance increase is not linear with the number of added resources and your limited to 8-socket systems!

Bull’s BCS solves these issues and shows that you can scale up beyond 8-socket systems without compromising performance. HPC technology delivered to x86 enterprise-class servers. Thanks to the Bull’s BCS eXternal Node-Controller!

In the next blog posts we will deep dive the BCS technology and uncover the secret sauce that makes the BCS sooo awesome!

Source: Bull, Intel, Wiki, Spec.org

 

About these ads

One thought on “Bull’s Implementation of a Glued Architecture

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s