BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Talk "Decoupled Compressed Caches" by Prof. David Wood\, Universit
 y of Wisconsin\, Madison (USA)
DTSTART:20140526T153000
DTEND:20140526T163000
DTSTAMP:20260407T113325Z
UID:c4651770d2957ec41a64ef31a7ea0d95ce5a245c1a7c76001510eefc
CATEGORIES:Conferences - Seminars
DESCRIPTION:Prof. David A. Wood has been a Professor in the Computer Scien
 ces and Electrical and Computer Engineering (by courtesy) Departments at 
 the University of Wisconsin\, Madison since 1990. Dr. Wood received a B.S.
  in Electrical Engineering and Computer Science (1981) and a Ph.D. in Com
 puter Science (1990)\, both at the University of California\, Berkeley.\n
 Dr. Wood was named an ACM Fellow (2005) and IEEE Fellow (2004)\, received 
 the University of Wisconsin's H.I. Romnes Faculty Fellowship (1999) and V
 ilas Associate (2011)\, and received the National Science Foundation's Pre
 sidential Young Investigator award (1991). Dr. Wood is Area Editor (Compu
 ter Systems) of ACM Transactions on Modeling and Computer Simulation\, is 
 Associate Editor of ACM Transactions on Architecture and Compiler Optimiz
 ation\, serves as Chair of ACM SIGARCH and ACM Council Representative\, se
 rved as Program Committee Chairman of ASPLOS-X (2002)\, and has served on
  numerous program committees. Dr. Wood is an ACM Fellow\, an IEEE Fellow\
 , and a member of the IEEE Computer Society. Dr. Wood has published over 7
 0 technical papers and is an inventor on thirteen U.S. and International 
 patents\, several of which have been licensed to industry.\nDr. Wood co-le
 ads the Wisconsin Multifacet project with Prof. Mark Hill (URL http://www
 .cs.wisc.edu/multifacet) which is exploring techniques for improving the 
 performance and energy efficiency of homogeneous and heterogeneous multipr
 ocessor servers.\nAbstract\nIn multicore processor systems\, last-level ca
 ches (LLCs) play a crucial role in reducing system energy by i) filtering 
 out expensive accesses to main memory and ii) reducing the time spent exe
 cuting in high-power states. Cache compression can increase effective cac
 he capacity and reduce misses\, improve performance\, and potentially redu
 ce system energy. However\, previous compressed cache designs have demons
 trated only limited benefits due to internal fragmentation and limited tag
 s.\nIn this talk\, I discuss our work on Decoupled Compressed Caches (DCC)
 \, which exploit spatial locality to improve both the performance and ene
 rgy-efficiency of cache compression. DCC uses decoupled super-blocks and n
 on-contiguous sub-block allocation to decrease tag overhead without incre
 asing internal fragmentation. Non-contiguous sub-blocks also eliminate the
  need for energy-expensive re-compaction when a block's size changes. Comp
 ared to earlier compressed caches\, DCC increases normalized effective cap
 acity to a maximum of 4 and an average of 2.2 for a wide range of workloa
 ds. A further optimized Co-DCC (Co-Compacted DCC) design improves the ave
 rage normalized effective capacity to 2.6 by co-compacting the compressed 
 blocks in a super-block. Our simulations show that DCC nearly doubles the
  benefits of previous compressed caches with similar area overhead. We als
 o demonstrate a practical DCC design based on a recent commercial LLC des
 ign.  
LOCATION:BC 410 https://plan.epfl.ch/?room==BC%20410
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR