You Can't Make Decisions Without ALL The Facts
What is it? A world class high performance tuning expert in a box.
PerfMiner is a tool that collects over 1000 elements of performance data on the nodes of your computing facility, operating in the background with near zero overhead, recording hardware performance counters, configuration and OS settings, all while your system is actually running its normal load. This is already revolutionary and delivers an unprecedented level of drill-down detail and discovery on performance problems. But we go further: We combine this with a second revolutionary step; using modern statistical data mining techniques to discover patterns and clusters to clearly identify poorly performing outliers, nodes, bottlenecks and configuration issues in your computational workload. These patterns feed into an expert system so we can link them to specific recommendations, fixes and actions. For example, in one application, run millions of times per day, we boosted performance 35% with one configuration setting.. PerfMiner began as an academic exercise in high performance computing a dozen years ago, and with continuous development since has scaled into a mature and powerful commercial suite ready to bring clarity and focus to your facility management decisions.
There are lots of dashboards out there. How is PerfMiner different?
Other tools display tons of metrics, graphs, pie-charts and gauges. They hope the customer knows what they mean. Some customers do; but in the modern computing facility there are literally hundreds of tuning points for the network, the nodes, the OS, independent software and more. It takes many years of experience to translate those displays into diagnoses of specific ailments, like knowing a recently recompiled app has inexplicably stopped using vectorization, or is no longer using cache effectively, or that some nodes have stopped pinning threads and are dragging the entire system down. These problems leave signature patterns in all that dashboard data, but a hundred dials moving every second overwhelms the human capacity for pattern recognition. That's where PerfMiner comes to your rescue; in the last ten years data mining has advanced strongly in various fields of marketing, genetics, and other big data ventures. The mountain of performance data that can be collected by PerfMiner opens a new frontier where we can apply the same data mining ideas, not to find a cluster of products bought by expectant parents or clusters of suspicious bank transactions, but to identify clusters of nodes in your facility that are mis-configured, and applications you run that are thrashing the cache.
TIME IS MONEY AND PERFORMANCE IS PROFIT
What did your facility cost? What if you could do twice as much with it? More computation often enables new paradigms in speculative design simulation, more finely grained modeling, or deeper searches in optimizations. With more computation your developers may gain the power to implement models previously beyond your grasp, so you can reap far more than just an incremental, proportional gain in results. Performance can be a game changer and the difference between success and failure. PerfMiner lets you do more with less staff, fewer consultants, less hardware, fewer developers, less downtime and fewer uncertainties. Analogous to anti-viral software, PerfMiner will identify problems virtually as they appear, which will minimize their impact on your facility. It can save you money directly and increase your computational capacity. That can increase your top line, and by decreasing your expenses and waste, provide a double lift for your bottom line.
Philip J. Mucci
President & Founder
Minimal Metrics was founded by Philip Mucci in 2012, with members of his previous company Samara Technology Group (founded in 2008) which consisted of a number of Philip's ex-colleagues from SiCortex. Known as the founder of PAPI, Philip has a Master's Degree in computer science from the University of Tennessee, Knoxville (under Jack Dongarra at the Innovative Computing Laboratory) and received his Bachelor's Degree from Johns Hopkins University. He is originally from Stow, Massachusetts and splits his time between Navarre, Florida and Kalmar, Sweden.
Dr. Anthony Castaldo
Chief Technology Officer
Dr. Anthony M. Castaldo has been working with Minimal Metrics since it's inception in 2012. Dr. Castaldo holds a PhD in Computer Science with an emphasis on high performance computing focused on computational linear algebra; he is an expert in statistical methods and data algorithms. He was previously a Research Assistant Professor at the University of Texas at San Antonio; dually appointed to the departments of Computer Science and Mechanical Engineering.
Graham Bates, Web Application Developer, is responsible for taking complex PerfMiner datasets and distilling them down to the most meaningful and clear visualisations via modern web development practices. Originally from Cambridge, UK, he now bounces between Amsterdam and London. Graham holds a Computer Science degree.
DR. JEFF DURACHTA
With over 26 years of experience in HPC platforms and environments and the applications that run in them, Jeff has spent his career working with end-users in the scientific, technical, and research communities. This work encompasses new HPC application development as well as legacy application migration in HPC environments. Jeff holds a PhD Additional experience includes extensive benchmark and performance analysis, workflow analysis and optimization, tool development, technology consulting, large system software design, development and implementation as well as project management. Clients, collaboration partners and employers include national laboratories associated with NOAA, DOE and DOD as well as vendors such as Cray, Intel, SiCortex and IBM.
Tushar Mohan has a formal background in computer science and has worked over a decade in the area of supercomputing. He has held performance engineering positions at IBM Research, Lawrence Berkeley National Laboratory, SiCortex and co-founded Samara with Philip. Tushar has a masters degree in computer science from the University of Utah and a bachelors degree from the prestigious IIT Delhi. Tushar is originally from New Delhi, India.
CHIEF MARKETING OFFICER
Rosalinda Mendez managed a multi-million dollar university research center for over a decade. During her tenure she oversaw the creation of a state of the art visualization facility and played and integral role in the creation of the research computing facilities for the Texas Learning and Computation Center. She held committee positions for the Coalition for Academic Scientific Computation and was named a Technology Fellow for the Center for Houston's Future. In addition, Rosalinda was the founder and CEO of NOA Research, LLC. Ms. Mendez holds dual bachelor's degrees from Vanderbilt University and a Masters degree from the University of Texas at Austin.