November 27, 2012
At SC12, Adaptive Computing announced its Moab HPC Suite 7.2 release, which includes several productivity enhancements and introduces support for Intel Xeon Phi coprocessors. The workload management vendor also launched two new products as part of its Moab HPC Suite: Application Portal edition, which adds support for a wider variety of applications, and Remote Visualization, a type of technical compute cloud.
Adaptive debuted their big Moab 7.0 release back in March and now they've come out with an incremental release, Moab HPC Suite 7.2. One of the major highlights of this version is support for the Intel Xeon Phi coprocessor, or "Intel's answer to the GPU," as company rep Chad Harrington put it when I sat down with him during SC12 in Salt Lake City.
Moab HPC Suite automatically detects installed Phi chips and determines how many cores are available. It also collects other metrics in real-time to enhance scheduling and optimization. It addresses questions such as: Is it hot? How much RAM is it using? Will it support additional workload or should workload be removed? Moab interacts with the coprocessor and manages it very efficiently, says Harrington.
Harrington told me that Adaptive customers who were beta testing the Phi chips have reported "great performance increases," and find the Phi easier to work with from a programming standpoint, compared to GPUs. While Adaptive also supports GPUs – including the latest graphics chips from NVIDIA and AMD – they are especially keen on the Intel Phi technology.
"With the introduction of the Intel Xeon Phi technology, we're seeing a new generation of supercomputers that are faster and more agile than ever," comments CEO Robert Clyde. "Adaptive is proud to offer Intel Xeon Phi capability in its latest version of Moab HPC Suite, to allow today's HPC centers to take full advantage of Intel Xeon Phi cores without the need for extensive reprogramming of their systems."
Another new Moab capability, one that was developed in response to customer requests, is dual-domain scheduling for Cray systems, which allows for a single job to straddle both Cray and non-Cray nodes. The Oak Ridge National Laboratory's Titan supercomputer, the current TOP500 chart topper, is an Adaptive customer who is using this heterogenous scheduling option.
The latest release also includes an upgrade to the Moab accounting and usage module, which is very cloud-like in its "pay-per-drink" model. Adaptive has added the ability to automate periodic budget resets as well as the ability to implement roll-over minutes – which means if you didn't use all of your allocation from last month, you can use it the following month.
Users of Moab 7.2 will want to take note of RPM-based deployments, a Linux-oriented package management system that minimizes installation time. A from-scratch installation, including downloading software, takes about eight minutes.
The Moab 7.2 release is already showing up in some very high-profile systems, for example, the COSMOS supercomputer, launched by Professor Stephen Hawking earlier this year. Housed at the University of Cambridge, the SGI UV 2000 is the most powerful shared-memory supercomputer in Europe, outfitted with 1,856 Intel Xeon E5 cores and 1,891 Intel Xeon Phi cores. As such, optimal scheduling and management are a top priority and will help the system fulfill its role in unlocking the mysteries of the universe.
"Research in fundamental cosmology is fast moving and internationally competitive," commented Professor Paul Shellard, COSMOS Director, in an official statement. "We have to adapt our flexible operating model rapidly, and we need a company breaking new ground to support the very latest HPC technologies, thus we selected Adaptive Computing for our workload management software."
As part of its SC12 news push, workload management specialists Adaptive Computing launched two new additions to its Moab HPC Suite: Application Portal Edition, which provides single-point access to common technical applications, and Remote Visualization Edition, which enables a technical compute cloud. The company reports these two new product versions "leverage next-generation access models to simplify the collection and interpretation of data, improving the time it takes to achieve meaningful results."
Application Portal Edition
Technical and engineering applications need to be able to integrate with the job scheduler, and this used to be a manual process that required significant HPC expertise. Adaptive has automated this functionality into a portal to allow users from all backgrounds to start their jobs, check statuses, and get results. Moab Application Portal Edition shifts the skill level from power users to novice users, Harrington explains. The portal, which was designed in collaboration with NICE Software, offers application-centric job submission templates for common applications in a variety of domains, including manufacturing, energy, life-sciences, education and others. The interface relies on NICE technology on the front-end for integration with the different applications, and Moab technology on the backend, for the scheduling and the sharing.
Remote Visualization Edition
A technical user that's doing simulation and modeling used to require an expensive workstation with a dedicated graphics processor, and data would need to be moved to the workstation in order to be processed. With remote visualization, all the compute-intensive work is happening in the datacenter or server room and only pixels are pushed to the remote site. This allows the company to save money on hardware and it's also faster and more secure because the data stays in the datacenter.
Remote visualization lets users around the world access and manipulate the same set of data. Harrington gives the example of a car manufacturer based in Germany who has built a vehicle simulation model and now their California lab wants to do some analysis. To ship a few terabytes of data from Germany to California is expensive and time-consuming, but this solution lets them do the visualization remotely in Germany and view the results from California over the Internet.
"In terms of bandwidth, pushing pixels is less bandwidth-intensive than your average Youtube video," notes Harrington. "If you try to move the data itself, it's not feasible, but a picture of the data works fine over a company's internal network and the consumer Internet," he adds.
Is this cloud? I ask.
"Cloud is about independence of data," Harrington responds. "It doesn't matter where the compute happens. It would even be possible to do this on an iPad. Adaptive calls this a technical compute cloud: the visualization happens somewhere else and you're witnessing it locally."
"This is along the same idea to VDI [Virtual desktop infrastructure], except in traditional VDI, you're using Microsoft Office or some kind of general productivity app, but in this case, you're using a simulation app, ANSYS Fluent or the like, technical HPC apps."
"The key here is that the processing is happening "elsewhere" – it could be in a different room in the same building, or in a public cloud like Amazon, or a company-owned datacenter on the other side of the world."
Adaptive developed this offering in partnership with NICE Software. As with the Moab Portal Edition, the NICE technology corresponds to the front-end, and they've combined it with the Moab scheduler, which works to manage the hardware side. A GPU can have hundreds of cores, explains Harrington, and Moab schedules the allocation of those cores. So a workload from User A may require 10 cores, while a workload from User B needs 30 cores and User C's workload wants 50 cores, and so on. Moab enables the sharing of one GPU across many users.
Earlier this month, Adaptive announced the 7.2 release for its Moab Cloud Suite. The Cloud Suite product comes with the same core Moab intelligence engine as the HPC suite, but offers specific features for private cloud. The latest version was designed for ease of integration to minimize the need for system upgrades. Other enhancements include multi-group management, a streamlined dashboard portal, and periodic budget reset capability.
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?