Parallel Databases Home Qiong Luo Hong Kong University

Parallel Databases @ Home Qiong Luo Hong Kong University of Science & Technology http: //www. cse. ust. hk/~luo

Parallel Databases n Future of high performance computing [De. Witt and Gray, CACM 1992] n n n Parallelism metrics: scaleup and speedup Parallel architectures: Shared-memory, shareddisk, shared-nothing Pipelined and partitioned parallelism n n Intra-operator parallelism: split and merge Specialized parallel operators All systems in this era ran in (super-)computer labs. Gongshow Qiong Luo @ CIDR 2007 2

Volunteer Computing @ Home n A bunch of distributed computing projects utilizing home PCs over the Internet (2000 -) n n n n SETI@home, 3 million users, TFLOPS-PFLOPS folding@home Einstein@home LHC@home Predictor@home Rosetta@home … All tasks are running on private computers at volunteers’ homes. Gongshow Qiong Luo @ CIDR 2007 3

Current Parallel Processors @ Home n CUDA n NVDIA Ge. Force 8800 video cards (Nov 06) n n n 16 SIMD multiprocessors, each of eight processors Over 300 GFLOPS (10 X Intel 3. 0 GHz Core 2 Duo) CBEA (The Cell Architecture by STI) n Sony Playstation 3 game console (Oct 06) n n One Power Processing Element (PPE) Eight Synergistic Processing Elements (SPEs) Commodity processors with massive parallel processing power Gongshow Qiong Luo @ CIDR 2007 4

Parallel Query Processing @ Home ? n n n There are probably applications for it. We might or might not need a full-fledged parallel database system. There will be a learning curve for the emerging hardware architectures. The @home computing paradigm requires us to rethink many issues. There is a wealth of literature on parallel. DB. Comments are welcome! Gongshow Qiong Luo @ CIDR 2007 5