微軟願景 – be what’s next 用 創新打造新世代的 it 藍圖
DESCRIPTION
微軟願景 – Be What’s Next 用 創新打造新世代的 IT 藍圖. 微軟亞洲研究院常務副院長 趙 峰. Microsoft Research (MSR). Cambridge, MA. Silicon Valley, CA. Redmond, WA. Cambridge, UK. Beijing, China. Bangalore, India. MSR-Asia @ 13. Founded Nov 5, 1998 200+ researchers 3000+ papers 260+ tech transfers. Big Data. - PowerPoint PPT PresentationTRANSCRIPT
Microsoft Research (MSR)
Cambridge, MA
Cambridge, UK
Silicon Valley, CA
Bangalore, India
Redmond, WA
Beijing, China
Advance the state-of-the-art in computing
Rapidly transfer technology to products
MSR-Asia @ 13• Founded Nov 5, 1998• 200+ researchers• 3000+ papers• 260+ tech transfers
Knowledge discovery in huge and heterogeneous data
Social structure
Traffic flows and road networks Points of interest User-generated content
Human mobilityUser locations
Knowledge Mined from Taxicabs
Recommending Driving Direction
Detecting Events
Connecting Taxi Drivers and PassengersUrban Planning
What is Index Serving?
“Microsoft Research”
http://research.microsoft.com/en-us/• Bing is serving billions of queries
every day• Each query searches for the best
webpages among billions of web documents
• It is the most critical infrastructure for the search engine
What is Tiger?• Tiger is the next generation index
serving platform for Bing• Leveraging Solid State Disks (SSD) to
deliver great performance improvements
• Translates to significant annual savings, and better user experience for Bing
• A close collaboration between Bing engineering teams and MSR Asia– Three researchers working almost full-time
for one and a half years among the product teams
– Demonstrates our commitment to leveraging research to improve Microsoft products
Data Center Connectivity• Challenges– Scaling: connect 100K-1M servers– Fault tolerance: failures are common– High performance: highly demanding apps – Low cost: commodity devices whenever
possible• Opportunity – Data centers owned by single org;
deployment can be easier than on Internet
ServerSwitch Design
Softw
are
Hard
war
e
PCI-E
• Packet forwarding in commodity switching ASIC– High performance and limited
programmability
• Low latency and high throughput interconnection
• Full programmability at server– Kernel module for low latency– User space for
programmability
NSDI’11 Best Paper Award
Kinect Identity
Face recognitionRobust to varying light condition
Fusion of multiple signaturesFace + clothes + body height
Handling real world variationsSignatures are learned from real data
User Identity
faceclothes height
+ +
Collaboration with Kinect Platform Team
Turning Ideas Into RealityKinect Contribution Highlights
• Kinect Identity: shipped Nov 2010
• Kinect Head Pose Tracking: shipped Aug 2011• Kinect-based Object Digitization: shipped June
2011• Avatar Kinect: shipped July 2011