CERCS Creating System Solutions for Future Technologies

1
CERCS - Creating System Solutions for Future
Technologies

• Karsten Schwan
• College of Computing
• Mustaq Ahamad, Doug Blough, Robert Butera, Andre
  Dos Santos, Richard Fujimoto, Mary Jean Harrold,
  Wenke Lee, Ling Liu, Ken Mackenzie, Peter
  Manolios, Vincent Mooney, Krishna Palem, Calton
  Pu, Kishore Ramachandran, Dave Schimmel, Olin
  Shivers, Yannis Smaragdakis, Jim Xu, Sudha
  Yalamanchili, Greg Eisenhauer, Steve Ferenci,
  Phil Hutto, Maria Hybinette, Thom McLean, Rajamar
  Krishnamurthy, Van Oleson, Kalyan Perumalla,
  George Riley, Matt Wolf, Neil Bright

2
Future Computing Applications

Ubiquitous Capture capture information
wherever/whenever useful Ubiquitous Access
interact with information/others with
ease Ubiquitous Interpretation use computers
and computational tools to make information and
interactions meaningful to end users Many Case
Studies Distributed Sensor Networks
Collaborative, Remote Science and Engineering
Aware Homes and ... Distributed Simulation,
Games, VEs
3
What is Systems Research?

• Creating new hardware and software technologies
• to enable future applications (previous talk)
• to create technological advances and take
  advantage of them
• to remove technological barriers
• in order to build integrated systems

4
Omnipresence of Linked Computing and Information
Devices Operational Information Systems
Airplane Data Traffic
Operational Flight Displays
Airport LAN
High Performance Computing Real-time Decision
Tools
FAA Flight Data
Real-Time Information Transport
capture, display, transport, filter, transform
Simulation
Optimization
Gate Readers
Equipment Inspection

Cluster Computing Real-Time Information Processing
Wide-area Transport
Passenger paging and response
Airport LAN
Crew and Equipment Status
Visualization
Real-time Situation Assessment
Baggage Displays
Scalable Robust Services
Storage
Baggage Status
Recovery and Replay
Security Systems
5
Advances, Expectations, Barriers

• Advance
• universal access to information and computation
• Expectation
• information delivery to the right places, in the
  right forms, at the right times
• Barrier
• appropriate deployment of distributed,
  heterogeneous, layered, dynamic resources
  (platform diversity)

6
CERCS Approach

• Integrated systems on programmable platforms
• Integrated systems
• Composable different networks and hardware, OS,
  middleware, languages separately programmed
  applications and components ...
• Safe and Secure multiple users separation
  across components
• Adaptive dealing with dynamic change in
  resources and needs
• Scalable, Robust, Available, Efficient in s,
  size, perception
• Programmable platforms
• Runtime extension, configuration, adaptation,
  specialization

7
I.a Composed Flow on Extensible Platform
C1
C2
C3
C4
D D D D D D D D D D D D D D D D D D D D D D D D D
D D D D D D
Programming Abstractions
(Middleware)
(Language Technologies)
C1
Application
C4
Operating System
C2
Co-processor
C3
Network
8
I.b Linking to the External World
9
I.c Composition, Extension, Partitioning
Sink
F( )
Source
Filter Function
if ((flight incoming to Salt Lake)
((flight outgoing from Salt Lake)) return
1 / submit into channel / return 0
/ suppress data /
10
II.d Runtime Adaptation for Performance
11
I.e Specialization for Power
Static Power Savings integer units in terms
of switching OFF of units
12
I.f Composition and Platform Extension
Application Component
Host
Host
Application Component
Middleware/OS
OS
OS
Extension Modules
Active NI
Active NI
Active Network
13
I.g Composition and Hardware Configuration
Control Steering Logic
BASE BLOCK
DECISION BLOCK
MUX
Winner
CTRL
BASE BLOCK
Loser
MUX
CTRL
One Cycle
DECISION BLOCK
BASE BLOCK
MUX
CTRL
Winner
BASE BLOCK
Loser
MUX
CTRL
14
Joint Vision I

• Integrated systems on open, shared platforms can
  substantially advance state of the art
• ability to utilize increasingly cheap and
  omnipresent computing/networking platforms
• scalability and availability through parallelism,
  distribution, and content-dependent operation
• improved performance and quality of service
  through runtime adaptation and specialization

15
Joint Vision II

• Integrated systems on open platforms offer
  substantial challenges (barriers)
• security and protection in open systems
• methods of effective runtime quality management
  for dynamic applications on dynamic platforms
  adaptation and specialization
• need for skills from many areas of CS
• languages, compilers, runtimes middleware and
  distributed systems communications and networks
  high performance, embedded, and real-time
  systems, ...

16
Center for Experimental Research in Computer
Systems (CERCS)

• Major advances in computing systems in the next
  decade and beyond will arise from integration of
  technologies to solve new problems and generate
  new capabilities.
• The role of the center is to promote such synergy
  via capabilities for experimentation.
• The center provides the infrastructure to support
  experimental efforts in the design of the future
  generations of computing systems and software.

17
Middleware for Heterogeneous/Embedded Systems
Calton Pu, Karsten Schwan, Greg Eisenhauer, Matt
Wolf, many Ph.D. students
18
Application 1Heterogeneous Portals Across the
Network
ECho
Downsampling
Channel G
Channel A
Channel C
ipaq Display
3D Visualization Steering
Channel D
Channel B
Residual Calculation
VisAD 3D Active Interface
Channel E
Channel F
2D Applet Visuali- zation
JECho
19
Application 2 Differential Processing to
ConserveLimited Wireless Bandwidth
20
Application 2 Details Sensor Stream in Wired
and Wireless Environments
Mobile Device

• Automatic adjustment of picture quality based on
  client needs and network capacity
• Rich set of controls (i.e. zoom, replay, )
• Deployment in larger scale settings

21
Solution Approach ECho/JECho

• Represent information flows as event streams
  (publish/subscribe model)
• Transparency of local and remote receivers via
  event channels, implemented with underlying
  peer-to-peer communications
• Efficient, fully typed binary data transmission,
  based on dynamically defined event formats
• Dynamic extension of existing formats and
  discovery/operation on format contents
  (reflection)
• Interoperate with CORBA and Java (JECho) via IDL
  and XML

22
Differential Stream Processing with ECho
Sink
F( )
Source
Event differentiation by windowing
if ((target within range) and ((affected
units within range)) return 1 / submit
into channel / return 0 / suppress
data /
23
Sink
Source-filtered Streams
F( )
Sink-filtered Stream
Source
24
ECho Fine-grained Event Differentiation

• Binary generation of filter code
• Dynamic deployment of filters
• Simultaneous control of multiple sources or
  differential control
• Control backchannels
• Meta-information
• about code
• current ad hoc
• better approach reflection and lightweight IR
• about event types
• event formats and format services
• XML and binary representations

25
JEChos Modulators/DeModulators
26
Solution (cont.) Infosphere/InfoPipes

• Infopipe Software for Information Flow
• Specification, Generation and Composition,
  Specialization, Adaptation
• Specialization of Infopipes
• Eliminate redundant processing

27
Solution (cont.) Infopipes

• Adaptation through microfeedback

28
Solution (cont.) Platform OS/Net Support

• Infosphere OS Principles for Quality Management
  (Pu) TCP-friendly communications, kernel-level
  micro-adaptation (i.e., CPU Scheduling), ...
• DEOS (Schwan) Quality Management in OS Kernels
  (Linux) ECalls, QSockets, DWCS Packet
  Scheduling, K-ECho, /dproc, dynamic kernel
  extension
• ASAN Mackenzie, Yalamanchili, Schimmel, Schwan
  focus on communication co-processors, incl.
  dynamic network extension

29
Future Directions

• Embedded Systems
• Power Management (with Mooney, Palem, Pande)
• Resilient Systems (with Ramachandran, Blough,
  Pande, ...)
• Information-secure systems (with Ahamad, Pu, Lee,
  dosSantos, ...)
• Wide area systems
• Portals on the Internet
• Access Grid

30
Related Efforts

• Stampede (Ramachandran - GT)
• Java work by Pande, Smaragdakis (J-Orchestra)
• Internet Computing, OpenCQ, XWrap (Liu)
• MITs Oxygen Project
• UWashs OneWorld
• Berkeley, CMU, Microsoft