BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20131119T223000Z
DTEND:20131120T000000Z
LOCATION:301/302/303
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: As we head to Exascale, the computational hardware will undergo a significant revamp toward low-energy usage.  General purpose computational cores might be augmented with heterogeneous accelerator cores.  Memory subsystems would move toward deeper multi-level caches, smaller cache-coherence domains and non-DRAM memories.  Networks will trend toward more cost-scalable and performance-localized setups.  How runtime systems can bridge the gap between such architectures and modern complex and dynamic HPC applications is an open question.  Would the answer be an evolutionary hybrid approach using MPI between nodes and an OpenMP or OpenACC-like model within the node?  Would the answer be a completely revolutionary dynamic runtime system that would rethink the computation and data movement model from the ground up?  This panel will be a half-and-half setup---the first half will be short "appetizer" presentations by experts discussing both evolutionary and revolutionary solutions; the second half will be an open debate among the participants.
SUMMARY:Exascale Runtime Systems
PRIORITY:3
END:VEVENT
END:VCALENDAR