Ph.D. Candidate George Bezerra Receives Excellence in Graduate Research Award

George Bezzera

Com­puter Sci­ence Ph.D. can­di­date George Bez­erra has received the 2012 UNM Chap­ter of Sigma Xi “Excel­lence in Grad­u­ate Research” Award. The UNM Chap­ter of Sigma Xi, an inter­na­tional, mul­ti­dis­ci­pli­nary research soci­ety, presents the Excel­lence in Grad­u­ate Research Award to encour­age and rec­og­nize the research per­formed by a doc­toral stu­dent near the end of his or her Ph.D. dissertation.

Bez­erra presents his research in a col­lo­quium, titled “Com­mu­ni­ca­tion Local­ity and Energy Con­sump­tion in Chip Multi-Processors,” Fri­day, June 15 at 10 a.m. in the Cen­ten­nial Engi­neer­ing Cen­ter Audi­to­rium. A recep­tion fol­lows at 11 a.m. in the Stamm Commons.

Bez­erra is grad­u­at­ing from UNM this sum­mer and start­ing a post­doc at MIT in the Fall. He works with mod­el­ing and opti­miza­tion of energy con­sump­tion in mod­ern com­puter archi­tec­tures, in par­tic­u­lar mul­ti­core chips with dozens of cores. He is also inter­ested the sim­i­lar­i­ties between power con­sump­tion in chips and metab­o­lism in bio­log­i­cal organ­isms as these sys­tems scale in size. He holds a Bach­e­lors degree in Elec­tri­cal Engi­neer­ing and a Mas­ters degree in Com­puter Engi­neer­ing both from the Uni­ver­sity of Camp­inas, Brazil.

Col­lo­quium Abstract
Per­for­mance increase of future com­puter archi­tec­tures will be dri­ven by the grow­ing par­al­lelism of multi-core chips and con­strained by power con­sump­tion. To take full advan­tage of the multi-core design, the com­mu­ni­ca­tion pat­terns of par­al­lel appli­ca­tions must be opti­mized through care­ful map­ping of data to cores, so that com­mu­ni­ca­tion dis­tances and energy con­sump­tion are minimized. We present a new method for data place­ment in Chip Multi-Processors (CMPs), which reduces on-chip energy con­sump­tion by tar­get­ing com­mu­ni­ca­tion local­ity and load-balancing. Our method is exact and can be solved in poly­no­mial time, improv­ing on ear­lier heuris­tic approaches that do not pro­vide guar­an­tees on solu­tion qual­ity. Sim­u­la­tions on a 64-core sys­tem showed an aver­age reduc­tion of dynamic energy con­sump­tion of 49.8 percent(and as much as 84.1 per­cent), with per­for­mance gains of up to 16.9% on par­al­lel sci­en­tific benchmarks.

Media con­tact: Tamara Williams (505) 277‑5859; email: tamara@unm.edu