Monday, June 7th, 13:00 CEST
Dr. Peter Kogge, University of Notre Dame, Indianna, USA
The Cambrian Explosion in Architecture – Are We “There” Yet?
The Cambrian period of life represented a time of explosion of life forms, all driven by larger collections of cells with increased specialization. Given the recent explosion in novel architectural features, it is natural to assert that we are at least entering a similar “golden age” of computing. This talk explores this assertion, with a comparison of advances in architecture with the evolution of life, and a conclusion that no, we are not quite at the Cambrian point yet, but the diversity we see now is a sign that architecture a generation from now may look radically different than today.
The main talking point is that evolution in architecture is driven by changes in what we want to compute, coupled with limitations in the available technology. In the past we have seen at least two such ``walls'' (memory and power) whose vanquishing required significant advances in architecture. In this talk we will discuss evidence on the emergence of a new third wall dealing with data locality (a site of computation is “here” and data is “over there”), which is prevalent in data intensive applications where computation is dominated by memory access and movement – not flops. Such apps exhibit large sets of often persistent data, with little reuse during any single computation, no predictable regularity, significantly different scaling characteristics, and where streaming is becoming important. Further, as we move to massively parallel algorithms running in the cloud, these issues will get even worse.
Solving such problems will take a new set of innovations in architecture to overcome. In addition to data on the new wall, the talk will look at several alternatives, and relate them to changes in life evolution.
One technique in particular, the concept of migrating threads, will be introduced, and related to cell functions that support various forms of mobility. The paper “Locality: The 3rd Wall and The Need for Innovation in Parallel Architectures” in the conference proceedings discusses this concept in detail, with this talk also suggesting other alternatives based on neuromorphic concepts.
PETER M. KOGGE is the McCourtney Professor of Computer Science and Engineering at the University of Notre Dame, a retired IBM Fellow, and a founder of Emu Solutions, now Lucata Inc. He is a fellow of both the IEEE and AAAS. His research interests are in massively parallel computing paradigms, processing in memory, and the relationship between massive non-numeric applications, emerging technology, and computer architectures. He holds over 40 patents and is author of three books, including the first text on pipelining and an upcoming text on models of computing. His Ph.D. thesis led to the Kogge-Stone adder used in many microprocessors. Other projects included the IOP - the world's second multi-threaded parallel processor which flew on every Space Shuttle, the IBM 3838 Array processor which was for a time the fastest floating point machine marketed by IBM, RTAIS and PIM Lite - systems with significant non-numeric computation built into a memory controller, and EXECUBE - probably the world's first multi-core processor and first processor fabbed on a DRAM chip. In 2008, he led DARPA's Exascale technology study group, which resulted in a widely referenced report on technologies and architectures for exascale computing. His startup, Emu Solutions, has demonstrated the first scalable system that utilizes mobile threads to attack large-scale big data and big graph problems. Dr. Kogge has received the Daniel Slotnick best paper award (1994), the IEEE/ACM Seymour Cray award for high performance computer engineering (2012), the IEEE Charles Babbage award for contributions to the evolution of massively parallel processing architectures (2014), the Gauss best paper award for high performance computers (Int. Supercomputing Conf. 2015), and the IEEE Computer Pioneer award (2015) (Highest award from IEEE Computer Society).