Entries with tag non-uniform memory access.

Techniques Enable Efficient Cloud Computing

Researchers from the University of California, San Diego, and Google have developed techniques that let hardware running cloud-computing resources operate more efficiently. Their approach uses data collected in real time on server clusters and works with non-uniform memory access (NUMA) technology, a memory approach for multiprocessing in which processors can access their own local memory faster than nonlocal memory. To make their technique work, the researchers apply a NUMA score to a process – in their experiments, they used Gmail backend server jobs -- to see how RAM is allocated in warehouse-scale computers and analyze performance impact. Optimizing the NUMA score, which can be accomplished by techniques such as clustering threads close to memory nodes, for example, could increase efficiency 15 to 20 percent. Google has already implemented the approach. The researchers presented their findings at the recent IEEE International Symposium on High Performance Computer Architecture in Shenzhen, China. (PhysOrg)(University of California, San Diego)(“Optimizing Google’s Warehouse Scale Computers: The NUMA Experience” @ University of California, San Diego) 

Showing 1 result.