July 11, 2013
Cloud-based high performance computing often comes as a result of purchasing clusters on something like Amazon's Cluster Compute Instances, where latencies are lower and parallelism is more easily achieved. Amazon already had a cloud computing service and realized there was a market to cater to applications requiring higher performance.
Another approach to supplying cloud-based HPC harkens back to the origins of Amazon’s cloud computing providing days. A company called Third Wave Systems (TWS) developed a new algorithm and high performance computing paradigm to solve machining simulation problems. In the same paper in which they introduced that method, they also outlined a plan to make such simulation capability available to those without the access to dedicated HPC resources.
“With the development of cloud computing,” their research states, “TWS is building the infrastructure for delivering AdvantEdge Software as a Service. In the SaaS model , AdvantEdge is hosted in the TWS HPC environment while customers remotely upload and submit AdvantEdge simulations via the Internet.”
In order for that to be possible, TWS must have an abundance of HPC resources that they do not always utilize. Much of the demand for cloud-based HPC is driven by the institutions who seek out burst capacity for their existing HPC clusters or those for whom buying those resources would not make sound financial sense.
These applications would be of the highest value to tool manufacturers. Below is a snapshot diagram of temperature variance for a drilling tool, as one can see by the temperature increase where a hypothetical drill would be contacting the surface matter.
That image details partially what TWS called ‘thermal-mechanical coupling’ and ‘contact correction,’ two features that are computationally intensive.
“The thermal-mechanical coupling and contact correction also entail a considerable amount of floating point operations and memory access cost,” the paper mentioned. “For a machining simulation with refined finite element mesh (>100,000 elements), the solution time running in sequential mode can take days or even weeks depending on the length of cut to be simulated. It is critical to reduce the solution time by improving the performance of the model”
Indeed, in order for their system to be a viable one to sell out as a SaaS, they had to make some tweaks to their algorithms such that it would run as a high performance system. According to TWS, they accomplished that by relying on OpenMP to power AdvatEdge, along with removing memory access bottlenecks in the context of thermal-mechanical coupling, as related by the research below:
“The data transfer for the thermal-mechanical coupling is found to be a performance bottleneck due to memory access. The implementation is optimized by coupling the two in a tighter manner such that the data exchange becomes more cache friendly. This change effectively removes the memory access bottleneck and significantly reduces the wall clock time for the operation.”
Overall, the HPC cluster consists of 23 12-core compute nodes, all with Intel Xeon X5680 processors along with four 40-core compute nodes of Intel Xeon E7-4870 multicore processors.
Their SaaS setup is relatively straightforward, drawing from application submissions over the internet. Once someone purchases time on the TWS system, they can, according to the report, run and track their applications through TWS’s web server. Per the report, “SaaS Web Server populates a SaaS Job Database, which keeps track of user accounts, company affiliations and simulation status, and the Web Server also allows users to track their simulations via querying into the aforementioned database. File Manager handles the file transfer between SaaS Client and TWS, which includes simulation input file upload and result file download.”
The setup that facilitates users’ interactions with the TWS HPC cluster is diagramed below.
Since security is of course always an issue, especially for manufacturers who would prefer to have their experiments and simulations kept to themselves, TWS plans to encrypt data transferring to its clusters through a secure socket layer mechanism.
This is certainly an interesting approach to providing access to HPC. TWS is covering a very specific use case in the sense that this operation is meant for manufacturers wanting to run simulations on optimizing their physical tools (like drills, etc.). Such simulations need only happen a few times over a product development’s lifecycle.
Genomic applications, which are data and computationally intensive, are a prime candidate to increase the scope of cloud-based HPC. The oil and gas industry, which would find a significant benefit in cheaply testing drill designs, generally have advanced tech setups due to their data-intensive nature. With that said, so does CERN and their adoption of cloud computing to foster scientific collaboration and to provide excess computing services is well documented.
It may be worth it for some companies to go with a cloud provider that specializes in exactly the applications they need. If this TWS SaaS experiment finds its targeted clientele and produces the high performance results that clientele is looking for in a secure fashion, it is conceivable that smaller but under-utilized HPC clusters could be rented out in a similar fashion.
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?