konecny-hpa-xmt

konecny-hpa-xmt - Introducing the Cray XMT Petr Konecny...

Info iconThis preview shows pages 1–8. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Introducing the Cray XMT Petr Konecny November 29 th 2007 November 07 Slide 2 Agenda Shared memory programming model Benefits/challenges/solutions Origins of the Cray XMT Cray XMT system architecture Cray XT infrastructure Cray Threadstorm processor Basic programming environment features Examples HPCC Random Access Breadth first search Rules of thumb Summary November 07 Slide 3 Shared memory model Benefits Uniform memory access Memory is distributed across all nodes No (need for) explicit message passing Productivity advantage over MPI Challenges Latency: time for a single operation Network bandwidth limits performance Legacy MPI codes November 07 Slide 4 Addressing shared memory challenges Latency Littles law: Parallelism is necessary ! Concurrency = Bandwidth * Latency e.g.: 800 MB/s, 2 s latency => 200 concurrent 64-bit word ops Need a lot of concurrency to maximize bandwidth Concurrency per thread (ILP, vector, SSE) => SPMD Many threads (MTA, XMT) => MPMD Network Bandwidth Provision lots of bandwidth ~1 GB/s per processor, ~5 GB/s per router on XMT Efficient for small messages Software controlled caching (registers, nearby memory) Eliminates cache coherency traffic Reduces network bandwidth November 07 Slide 5 rigins of the Cray XMT Multithreaded Architecture (MTA) Shared memory programming model Thread level parallelism Lightweight synchronization Cray XT Infrastructure Scalable I/O, HSS, Support Network efficient for small messages Cray XMT (a.k.a. Eldorado) Upgrade Opteron to Threadstorm November 07 Slide 6 Cray XMT System Architecture MTK Linux Compute Service & IO RAID Controllers Network PCI-X 10 GigE Fiber Channel PCI-X Service Partition Linux OS Specialized Linux nodes Login PEs IO Server PEs Network Server PEs FS Metadata Server PEs Database Server PEs Compute Partition MTK (BSD) November 07 Slide 7 Cray XMT Speeds and feeds Threadstorm...
View Full Document

This note was uploaded on 12/27/2011 for the course CMPSC 240A taught by Professor Gilbert during the Fall '09 term at UCSB.

Page1 / 20

konecny-hpa-xmt - Introducing the Cray XMT Petr Konecny...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online