Tailoring Computer Systems for Big Data and Tiny Technology Nodes

Sadržaj:

The unprecedented rate of data collection and the wide use of machine learning techniques have created a need for efficient processing, movement, and storage of vast amounts of data. Unfortunately, the capabilities of today’s computer systems to meet such needs are being challenged by the end of the conventional technology scaling (a.k.a. the death of Moore's law). Moreover, the end of technology scaling and the emergence of new and radically different technologies threaten to disrupt the half a century old balance of system resources in unpredictable ways, rendering traditional designs of systems and applications suboptimal and obsolete. In this talk I will show how computation, communication, and storage in datacenters can be specialized to provide significant efficiency gains, and how the proposed specialization of individual system components  affects the balance of the entire system. I will then present my vision of a framework for optimal designs of systems and applications in the rapidly changing technology space.

Biografija predavača:

Đorđe Jevđić je završio Elektrotehnički fakultet u Beogradu kao student generacije na Katedri za Računarsku tehniku i informatiku. Potom je završio doktorske studije na EPFL-u, u Švajcarskoj, gde je radio na specijalizovanim arhitekturama računara i memorijskim sistemima za servere, kao i
post-doktorske studije na Univerzitetu u Vašingtonu u Sijetlu, gde je radio na konceptu aproksimativnog čuvanja podataka. Nakon toga je proveo godinu dana kao istraživač u kompaniji Microsoft Research u Redmondu, gde je radio na arhiviranju podataka u sintetički DNK. Trenutno je docent na Računarskom Fakultetu Nacionalnog Univerziteta u Singapuru. Dobitnik je više priznanja, uključujući nagradu za najbolji rad na konferenciji ASPLOS, IEEE Micro Top Picks, Intel Doctoral Student Honor Fellowship, i stipendije za postdoktorske studije Švajcarskog Nacionalnog Fonda za Nauku.