As the high-performance computing (HPC) community prepares to descend on Denver for SC17 next week, its members will arrive in the Mile-High City with more baggage than the usual rolling carry-on. They’ll also be packing some long-held expectations. One of these is that it’s more or less impossible to create a real HPC system—a massive single system image—in the cloud. I fully anticipate they will leave Denver with the opposite expectation.
That’s because SC17 attendees stopping by the Oracle Booth (No. 1101) in the Colorado Convention Center will see how nearly anyone can build HPC-class systems in the Oracle Cloud in just a few minutes, and with just a few clicks. We’re talking dozens of terabytes of memory and hundreds of cores, all presented as a single system and operating under a single instance of Linux.
And it’s not just easy to set up, it’s simple to write and run your programs since it’s not a cluster. It’s a Software-Defined Server, which combines multiple commodity servers into a single, high-performance system right-sized to fit any large enterprise workload. This is all enabled by TidalScale’s inverse hypervisor technology. Unlike traditional hypervisors that run multiple images on a single server, TidalScale’s inverse hypervisor technology enables Software-Defined Servers to combine the resources of multiple commodity systems – including cores, memory, storage and network – into a single system image. When you create a Software-Defined Server, all those resources are available to you, just as if they were resident on a single, Unified Memory Architecture (UMA) system. In fact, it looks like a single massive piece of hardware with as much memory, cores, and I/O as you need.
Last month, TidalScale announced Oracle Cloud integration at Oracle OpenWorld. Oracle Cloud provides compute, network and storage infrastructure on demand. With TidalScale in the Oracle Cloud, users will be able to use a simple point-and-click control panel to select from available Oracle bare metal cloud systems to configure a single Software-Defined Server with all the cores, memory, storage and network needed to handle their workload. When the job is done, users can reconfigure those resources into one or more systems to handle the next task. Using Oracle’s low latency 25G cloud network layer, we’re able to assemble dozens of nodes from anywhere in the Oracle Cloud Infrastructure.
The Computer You Need, Right-Sized for Your Problem, Rented by the Hour
Now, these won’t be the largest systems demonstrated at SC17, where the biggest bragging rights are earned with high Linpack numbers driven by systems with millions of cores. But for enterprises struggling to accommodate Big Data analytics and modeling problems—and not the scientific Grand Challenges that capture headlines in the supercomputing world—a single system with 10, 20, 30, 40TB or more of memory solves a lot of problems. Literally.
What’s particularly beneficial is that with TidalScale on Oracle Cloud Infrastructure, you'll be able to right-size the server to fit your workload. Need 20TB? No problem. 120 cores? You got it. 6TB and 156 cores? It’s all yours. And best of all, you can rent it by the hour.
This is what TidalScale on Oracle Cloud Infrastructure makes possible. At SC17, we'll show you how you can use TidalScale's WaveRunner™ control panel to quickly pull together up to five hefty Oracle bare metal servers into a single system. You can then visually monitor the performance and utilization of those nodes, as well as any storage or network associated with them. And you can reassign those resources to other tasks, entirely on the fly and in just a few clicks.
The Largest On-Demand, Shared-Memory Supercomputer in the World
But all this is just a precursor to what you can do later this year. Come December, a new release of TidalScale WaveRunner will enable you to configure the largest on-demand, shared-memory supercomputer in the world in configurations ranging up to 50TB of DRAM. (That will be more than 10 times larger than the next largest largest cloud servers, including some recently announced 4TB cloud instances that have been generating all kinds of buzz.) This means that for the first time, a large-scale HPC-class shared-memory application can be run on standard cloud infrastructure.
So while you’re at SC17, be sure to drop by Oracle’s Booth No. 1101. You may arrive with some old expectations, but I’m confident you’ll return home with new ones that are likely to have a major impact on how you tackle your next Big Data problem.