Sun Microsystems was an innovator and a leader in high performance computing from the onset of SMP-based servers and powerful workstations. That began to change with the introduction of “LINTEL”- clusters (Linux and Intel X86 servers) over a decade ago. While hindsight can indicate that mistakes – or misjudgments – were clearly made, perhaps on par with Research in Motion’s co-CEO’s now famous dismissal of the iPhone. At the time of the iPhone’s introduction RIM’s co-CEO saw it as “one more entrant into an already very busy space with lots of choice for consumers.” But was the iPhone a threat to RIM? “I would think that’s overstating it,” Balsillie quipped. It would seem that history does repeat – Sun was originally dismissive of Linux, among other technologies.
Once a very active partner and participant in the HPC world the Sun presence disappeared after Oracle completed its acquisition in 2010. With this disappearance a significant gap was created in the support of customers and further development of Sun HPC products. This presented significant risk to customers who faced an uncertain future and potential problems in their production clusters. Grid Engine, Sun’s most famous and popular HPC software product, at the time was deployed in over 10,000 data centers and used in ways that touch the lives of people across the globe. From scheduling the crew for every flight of the world’s leading airlines, to perfecting the accelerometers used in smart phones, designing F1 race car technology or German sports cars. Grid Engine is a staple in the research and development of many of the most recognizable products companies in the world.
The concern was real. Paul Brenner, PhD, P.E. at the Center for Research Computing, The University of Notre Dame shared “Grid Engine is a key piece of technology in our infrastructure. We were concerned about suitable support and stability for Grid Engine moving into the future. Univa Grid Engine helps us answer those concerns…”
More worrisome was the reality that Grid Engine was open source but did not have many contributions from a wide network of open source developers. The development model (and free price) relied entirely on Sun-paid engineers since adding features in the scheduler (the heart of the system) was not simple and presented quality issues for anyone outside a small core group of engineers.
According to Ernst Bablick, a 13 year veteran developer of Grid Engine, “I am working….in the team of engineers that is continuously improving and enhancing Grid Engine. Nevertheless I am often astonished how difficult and long-lasting it can be to design and implement new features.”
This is the reason Univa “airlifted” the core development team out of Oracle.
The Future
Since January 2011, Univa has made substantial progress improving Grid Engine and getting the product ready for the rapidly changing future which includes hyper-scale core counts, NUMA-based systems, GPUs, massive data processing, new architectures that include Cloud computing and scheduling mixed workloads that include Hadoop and other Big Data tools. Several customers are already clipping 100,000 cores clusters and others have integrated Hadoop with ease thanks to improvements over the open source version.
Univa Grid Engine 8.1 brings several new features in the scheduler that adds performance or reduces the support burden of production clusters. A list of features is available on the website here. A more comprehensive description has been blogged by the engineers who wrote the code. The blogs can be found here.
After Univa Grid Engine 8.1 ships, the beta program for the upcoming License Orchestrator product will begin. Univa License Orchestrator dynamically optimizes expensive third party software licenses. In some industries the cost of the applications used in the cluster is an order of magnitude above the cost of the infrastructure. Improving license use by even 1% could be worth as much as a million dollars on an annual basis. The beta program is limited. Interested organizations should sign up here.
Then there’s Big Data. For customers, product development becomes a more competitive and higher stakes game each year. An organization’s Big Data is the key to customer product design superiority and applying Big Compute to Big Data is key to timely product innovation, development and delivery. One can expect to see leaps and bounds in this subject as more people seek to add Hadoop into the system rather than building silos.
Information on Univa Grid Engine, the roadmap and the new release of 8.1 can be found here