Minnesota Systems Cloud Research Vision Jon Weissman Abhishek Chandra Distributed Computing Systems Group Department of CS&E University of Minnesota NSF.

Minnesota Systems Cloud Research Vision

Jon WeissmanAbhishek Chandra

Distributed Computing Systems GroupDepartment of CS&E

University of Minnesota

NSF Science of Cloud Computing Workshop, March 2011

Introduction: The Cloud Today

• Dominant Usage Modes– batch: analytics– hosting: web services– storage: archive/backup/sharing

end-user-neutral• Dominant Platform Modes

– high latency: install and access– limited distribution: few data-centers

localized

Analytics

Resultsout

Datain

with thanks to Ian Foster

Computation

Cloud Limitations: localized

• Large volumes of widely distributed data– too expensive to move PBs of data centrally– poor locality to data sources

• High latency deployment and access– limits highly network-sensitive user-facing services – limits short-term services

Þ in-situ/distributed, lightweight

Idea

• Make the cloud more “distributed”– “move” it closer to data– “move” it closer to end-users– “move” it closer to other clouds

• Make it lower latency– non-virtualized, on-demand

Example: Dispersed-Data-Intensive Services

blog1 blog2

blog3

Data is geographically distributed Costly, inefficient to move to central location

Nebula: A New Cloud Model

• Stretch the cloud– exploit the rich collection of edge computers – volunteers (P2P, @home), commercial (CDNs)

NebulaCentral

Nebula• Decentralized, less-managed cloud

– dispersed storage/compute resources– low latency deployment: native client

Example: Dispersed-Data-Intensive Services

blog1 blog2

blog3

Data is geographically distributed Costly, inefficient to move to central location

Challenges

• Algorithmic/systems challenges

• Organization drivers– CDN vs. volunteers– trusted local clouds?

• Vision paper: HotCloud 2009, DIDC 2011

Cloud Limitations: user-neutral

• Mobile users/applications: phones, tablets– resource limited: power, CPU, memory– applications are becoming ^ sophisticated

• Improve mobile user experience– performance, reliability, fidelity– tap into the cloud based on current resource state,

preferences, interests=> user-centric cloud processing

Cloud Mobile Opportunity

• Dynamic outsourcing– move computation, data to the cloud dynamically

• User context– exploit user behavior to pre-fetch, pre-compute, cache

• Multi-user sharing– Implicit sharing based on interests, social ties

Example 1

• Outsourcing– local data capture + cloud processing– images/video, speech, digital design, aug. reality

Server Server Server Server Server

Proxy

Code repository

….

….

Mobile end

Application Profiler

Outsourcing Client

Outsourcing Controller

Nebula could also be the back-end

Commercial cloud

Experimental Results -Image Processing

• Response time– Both WIFI & 3G– Up to 27× speedup– 219K, WIFI

• Power consumption– Save up to 9× times– 219K, WIFI

14

Avg. Time

Avg. PowerFace recognition

Example 2• Dynamic user profile

– contains activities in time and space– “read nytimes.com at 9am on the train; likes

technology articles”• Patterns are relationships between activities

– repetitive, sequential, concurrent, time-bounded– “user always does X and then does Y”

• Exploiting patterns: pre-fetching, pre-computing, caching in the cloud

User-centric cloud

RAi knows user i profile

Vision paper: University of Minnesota, CSE TR-11-006, March 2011.

Summary• Trends

– Dynamic large distributed data– Mobile users

• Our vision of the (a?) Cloud – locality of users, data– deep mobile integration, user-centricity

Thank you! Questions?