Slidecasts: Using the next-generation Cray XMT (uRiKA) for Large Scale Data Analytics

CSCS, the Swiss National Supercomputing Centre organized in May 2012 a course on “Using the next-generation Cray XMT (uRiKA) for Large Scale Data Analytics”. Here are now published the slidecasts of the different presentations.

Mario Valle (CSCS): Introduction to the workshop “Using the next-generation Cray XMT (uRiKA) for Large Scale Data Analytics”

Many critical Big Data problems are based on graphs. Unfortunately current Big Data approaches result in low performance on graphs since graphs are hard to partition across cluster nodes, are non-deterministic, and are highly dynamic.

The Cray uRiKA graph appliance addresses the challenge of delivering insightful analytics on graphs, not only in terms of its ability to handle size and complexity of relationships, but also in terms of its response time and speed of processing.

First goal of the workshop is to familiarize potential users to the functionalities offered by this machine, that is part of the CSCS portfolio.

James Maltby (Cray): Cray XMT/uRiKA overview

An overview of the XMT machine, now rechristened uRiKA, and the strategies it implements to effectively solve problems with irregular structure like graph algorithms.

http://youtu.be/YYLjchgpjpY

James Maltby (Cray): uRiKA and Graph Analytics

The uRiKA machine comes with a software stack specialized for graph queries. Through SPARQL statements the user could submit complex queries to obtain insightful analytics on graphs.

James Maltby (Cray): Cray XMT Multithreated programming model

The XMT heavily multi-threaded processor with its lightweight synchronization support using full/empty bits on all memory requires a different programming model to exploit all its power.

John Feo (Pacific Northwest National Laboratory): Programming the Cray XMT

In depth view of the XMT programming from people that use it for their research.

John Feo (Pacific Northwest National Laboratory): Dataflow on Cray XMT

How a classical problem can be redefined to exploit the XMT peculiar parallel architecture and memory structure.