Issue link: http://osercommunicationsgroup.uberflip.com/i/412884
S u p e r C o m p u te r S h o w D a i l y 9 Tu e s d a y, N o ve m b e r 1 8 , 2 0 1 4 I mentioned, we have experienced tremendous organic growth over the past few years, due in no small part to the unique customer experience we pro- vide and the relationships we build with customers as a result. I expect that this growth will continue into 2015. SCSD: What's most exciting for Silicon Mechanics in terms of attending SC14? EC: Everything! It's awe-inspiring to watch the next generation of supercom- puting professionals develop their skills in front of your eyes in the Student Cluster Competition. In addition, we believe that high performance computing is an important part of our future, and so it is helpful not just to show off our own capabilities, but also to get a sense of the state of the industry. Visit Silicon Mechanics at booth #2623. For more information, go to www .siliconmechanics.com, call 425-424-0000 or email info@siliconmechanics.com. Silicon Mechanics ( Cont'd. from p. 1) The project will provide astronomers with an unprecedented view into the universe and will share the data being generated through CANARIE, Canada's National Research and Education Network. Like any piece of existing infrastruc- ture, the network must be maintained and upgraded periodically, preferably before problems arise. The proliferation of the cloud and newly emerging bandwidth intensive cloud services combined with exponentially growing data sets are a cause of concern for R&E network oper- ators. Many of the networks in place today were not designed to carry terabits of traffic, let alone manage the traffic in a reliable fashion. Currently, the R&E community is working to ensure quality measuring, monitoring and controlling of network traffic, and this is where high- performance converged packet-optical technologies combined with intelligent software can help. Converged packet optical and soft- ware solutions can not only provide the necessary capacity to support R&E proj- Ciena ( Cont'd. from p. 1) ects like the SKA Project, but also trans- form the network to be more agile, pro- grammable and fully capable of support- ing the transport of big data sets as well as a plethora of specialized network serv- ices, "on-demand." SDN and NFV tech- nologies will be able to address some of these issues. For example, SDN and advanced technologies like real-time analytics software applications enable researchers to trial new research method- ology on a fully operational network without having to build a unique infra- structure for every use case – or in other words – they will have a network that they can leverage "on-demand." We are not there yet, but are within reach. The networks of the research and education community have always been a proving ground for evolving technolo- gies, and it is once again time to lead the world in advancing the capabilities of the network to ensure that collaboration and scientific discovery will not be stifled by the limitations of existing technologies. For more information, visit Ciena at booth #3315. "gateway" projects that the open source community has ever produced, because they are equally useful to Windows, Macintosh, GNU/Linux, Unix and virtual- ization users. Everyone needs a reliable way to store their files and build their networks. You can find out more about FreeNAS at the iXsystems booth, #2522. SCSD: What role do FreeBSD and OpenBSD play? MD: While FreeBSD is inside both FreeNAS and pfSense, you want FreeBSD and OpenBSD for maximum flexibility and features. Together, they provide exemplary networking, storage and security capabilities and run on any- thing from embedded devices through the largest Intel and AMD servers. SCSD: What role does bhyve play? MD: We all know that a handful of open source and proprietary hypervisors dom- inate the market with each having various baggage from their origins. bhyve does away with baggage like BIOS, CPU and device emulation, instead relying on the hardware virtualization assistance found in modern CPUs. The sheer elegance of bhyve allows it to stay out of the way while prototyping and deploying BSD Unix and GNU/Linux virtual systems. SCSD: Do you have any unique products G ainframe ( Cont'd. from p. 1) or services? MD: OpenZFS-backed NTFS and Linux network and virtualized block devices are proving very popular, and the appli- cations are endless. Every computing system deserves a proper backing store like OpenZFS with its integrated check- summing, snapshotting and redundancy, but the reality is that the world runs on NTFS and other traditional filesystems. Our block device and file-based systems give LAN and NOC administrators the "machine room" level of control and monitoring they have gradually lost, hence company name playing off "main- frame." SCSD: Any final observations? MD: The Supercomputing organizers said that the ratio of women to men has been about 50/50 for the technical ses- sions in recent years. This is exempla- ry in technical circles, and I wish to applaud the balance they have achieved. I hope the open source com- munity will also achieve this ratio sooner rather than later. Please look for opportunities to reach out to minorities of all genders, orientations, races and ages. Visit booth #3461 to learn more about bhyve, BSD Unix, the Open Source Initiative and Gainframe. For more information, go to www .gainframe.com, call 503-789-8978 or email sales@gainframe.com. image system, and each node can have multiple processor cores. AMD proces- sors can address 256TBytes of data, and this does limit the total memory space of the systems. A directory-based cache coherence protocol handles scaling, with significant numbers of nodes sharing data to avoid overloading the intercon- nect between nodes with coherency traf- fic, which would seriously reduce real data throughput. Basic ring topology with distributed switching allows for a number of differ- ent interconnect configurations that are more scalable than those provided by most other interconnect switch fabrics. Ring topology also eliminates the need for a centralized switch and includes inherent redundancy for multidimension- al topologies. The topologies used are two- and three-dimensional topologies (torus) that have the advantage of built-in redundancy, as opposed to systems based on centralized switches, where the switch Numascale ( Cont'd. from p. 1) represents a single point of failure. Distributed switching reduces the cost of the system because there is no extra switch hardware to pay for. It also reduces the amount of rack space required to hold the system, as well as the power consumption and heat dissipation from the switch hardware, and the asso- ciated energy loss of the power supply. Shared memory and OS simplify parallelization tasks. Running a single- image standard OS is an advantage for reliability, operations and system man- agement. The hardware integrates seam- lessly with the processor cache system and takes advantage of standard opti- mization techniques. NumaConnect provides an afford- able solution by delivering all the advan- tages of expensive shared memory com- puting for a cluster price point. Visit Numascale at booth #1923. For more information, go to www .numascale.com, call 832-470-8200 or email ts@numascale.com. THREE PRESENTATIONS TO NOTE ON YOUR SCHEDULE The Supercomputing 2014 workshop schedule for Tuesday, November 18 includes a presentation by Oracle Corporation's Christine Rogers, who will be talking about moving, managing and storing data in a presentation titled, "Why Archive? Drive Down Costs and Improve Efficiencies." Data growth, regulations, content dig- itization, lean budgets, and the need for storage efficiencies are all complicating the operational and financial pictures of enter- prises, which must store and protect infor- mation for years. While data retention chal- lenges are growing, organizations are also looking to leverage their data to gain insight into and grow their businesses. Learn how introducing an archive strategy for long-term data retention can help your organization drive down storage costs and improve operational efficiencies. Your organization can, then, focus on unlocking the value of your data for new business opportunities, according to Rogers' abstract for the talk. The session is scheduled for 10:30 to 11 a.m. in room 291. At the same time in room 388-390, you can hear the thoughts of Jidong Zhai, Jianfei Hu, Xiongchao Tang, Xiaosong Ma, Wenguang Chen on performance measure- ment with their presentation of their paper titled "CYPRESS: Combining Static and Dynamic Analysis for Top-Down Communication Trace Compression," a best paper finalist for the conference. According to the abstract, "Communication traces are increasingly important, both for parallel applications' performance analy- sis/optimization, and for designing next- generation HPC systems. Meanwhile, the problem size and the execution scale on supercomputers keep growing, producing prohibitive volume of communication traces. To reduce the size of communication traces, existing dynamic compression meth- ods introduce large compression overhead with the job scale." The authors propose a hybrid static-dynamic method that lever- ages information acquired from static analy- sis to facilitate more effective and efficient dynamic trace compression. This methodol- ogy, CYPRESS, extracts a program com- munication structure tree at compile time using inter-procedural analysis. This tree naturally contains crucial iterative comput- ing features such as the loop structure, allowing subsequent runtime compression to fill in, in a top-down manner, event details into the known communication tem- plate. Results show that CYPRESS reduces intra-process and inter-process compression overhead up to 5x and 9x respectively over state-of-the-art dynamic methods, while only introducing very low compiling over- head. The session will be chaired by Shirley Moore of the University of Texas El Paso. In the New Orleans Theater at 10:30 on Tuesday, Masoud Mohseni of Google will be speaking on the topic of Quantum Computing Paradigms for Probabilistic Inference and Optimization. Mohseni posits that over the past 30 years, several compu- tational paradigms have been developed based on the premise that the laws of quan- tum mechanics could provide radically new and more powerful methods of information processing. One of these approaches is to encode the solution of a computational problem into the ground state of a program- mable many-body quantum Hamiltonian system. Mohseni's talk will provide an overview of quantum computing paradigms and discuss the progress at the Google Quantum Artificial Intelligence Lab towards developing the general theory and overcoming practical limitations. He'll also newly developed algorithms known as Quantum Principal Component Analysis and Quantum Boltzmann Machine.