Here’s a collection of highlights, selected totally subjectively, from this week’s HPC news stream as reported at insideHPC.com and HPCwire.

>>10 words and a link

Sao Paulo’s center for HPC expands SGI installation;
http://insidehpc.com/2008/03/14/cenapad-sp-expands-sgi-altix-install/

New infrastructure-as-a-service conference on the horizon;
http://insidehpc.com/2008/03/17/new-infrastructure-as-a-service-conference/

SiCortex moves in at U Colorado’s Laboratory of Atmospheric and Space Physics;
http://insidehpc.com/2008/03/18/university-of-colorado-selects-sicortex/

SGI revs ICE Blades;
http://insidehpc.com/2008/03/19/sgi-announces-enhancements-to-altix-ice-blade-platform/

Multicore software workshop in conjunction with ICSE;
http://www.supercomputingonline.com/article.php?sid=15302

Sun releases SAM/QFS as open source;
http://insidehpc.com/2008/03/20/sam-and-qfs-available-as-open-source/

>>NEC claims 10-petaflop supercomputing breakthrough

NEC and the Tokyo Institute of Technology are claiming they have developed technology that will allow supercomputers to breach 10 petaflops. Apparently, they have laid the groundwork using a network of optical interconnections between nests of chips. The optical interconnects allow systems to perform chip-to-chip communication at 25 Gbps (no mention of latency).

The prototype converts electrical signals into optical signals using laser diodes, says our man at the Nikkei, and its connector bundles 1,000 fibres together to bring together the world’s most powerful aggregation of neighbouring chips.

The Japanese government says it could be ready by 2010. Will this technology power the next Earth Simulator, or is this simply a pipe dream? Frankly, it’s probably too early to tell. In the mean time, read the full article at http://www.theinquirer.net/gb/inquirer/news/2008/03/20/nec-super-computer-petaflop.

>>Cluster Resources announces new products

Cluster Resources, Inc. has made two product announcements at the Novell BrainShare conference in Salt Lake City this week. The first includes details and demos of the Moab Cluster Builder for SUSE Linux. Cluster Builder is pegged as being a turn-key cluster deployment and management solution for customers running SUSE Linux. Cluster Builder also incorporates the Moab Cluster Suite. This addition rounds out the overall feature set with a Web-based job submission portal, a graphical Windows-like administration interface and professional reporting capabilites.

The second announcement gives way to the Moab Hybrid Cluster product. Hybrid Cluster is an HPC solution that allows one to dynamically change cluster nodes between Windows CCS and SUSE Linux based on workload and application requirements. The Hybrid Cluster product is the result of a joint effort between Novell and Microsoft to create a unified cluster environment across both companies’ operating systems. Michael Jackson, president of Cluster Resources, will be on hand Tuesday and Friday of the event in order to co-present a talk entitled, “Adaptive High Performance Computing with SUSE Linux and Windows.”

For links to more info, see the post at insideHPC.com: http://insidehpc.com/2008/03/17/cluster-resources-product-announcements-at-novells-brainshare/.

>>An end to the dominance of relational databases

Bill McColl has a post (http://www.computingatscale.com/?p=46) over at his blog commenting on the end of the era of the RDBMS for storing and, more importantly, analyzing large stores of data. In the post he discusses some of the major alternative data projects and makes this observation:

Like the H-Store team, I too advocate that we get away from the legacy grip of SQL and instead start using much simpler and more powerful “little languages”, derived from mainstream programming languages, as the basis for our processing of both stored data and live data. Well designed domain-specific embedded languages (DSELs) will enable us to provide users not only with cleaner abstractions that are easier to learn and to use effectively, but also with implementations that are much more scalable, and higher performance.

This post is timely in that I just interviewed the folks at Pervasive software on their DataRush product, an interesting effort to create a portable, very high performance data analytics engine outside of an RDBMS by exploiting single-socket (really single address space) parallelism. Results so far are impressive. Look for an article in HPCwire over the next few weeks with more details.