Upload
sara-watts
View
218
Download
0
Tags:
Embed Size (px)
Citation preview
User and Network Interplay in Internet Telemicroscopy
Prasad Calyam (Presenter)Nathan Howes, Mark Haffner, Abdul Kalash
Ohio Supercomputer Center, The Ohio State University
IMMERSCOM, October 11th 2007
2
Topics of Discussion
Telemicroscopy Overview Motivation Use-cases Solutions
Telemicroscopy Session Model User and Network Interplay
Testbed for Experiments to Characterize Model Parameters Performance Analysis OSC’s Remote Instrumentation Collaboration Environment (RICE)
Features Demo Video
Conclusion
3
Telemicroscopy Overview Academia and Industry use computer-controlled scientific instruments
Electron Microscopes, NMR, Raman Spectrometers, Nuclear Accelerator For research and training purposes
Cancer Cure, Material Science, Nanotechnology
Instruments are expensive ($450K - $ 4Million) and need dedicated staff to maintain
+) Remote instrumentation benefits Access to users who cannot afford to buy instruments Return on Investment (ROI) for instrument labs Avoids duplication of instrument investments for funding agencies (NSF, OBOR) Useful when physical presence of humans around sample is undesirable
-) Remote instrumentation drawbacks Improper operation can cause physical damages that are expensive to repair
Telemicroscopy is remote instrumentation of electron microscopes
4
Telemicroscopy Use-cases Tele-observation versus Tele-operation
5
Telemicroscopy Solutions Hardware-based: KVM over IP (KVMoIP)
Encoder-Decoder pair for frame-differencing based video image transfers Pros: High quality video and optimal response times Cons: Expensive, Special hardware and high-end bandwidth requirements
Software-based: VNC – remote desktop software Raw or copy-rectangle or JPEG/MPEG encoded video image transfers Pros: Inexpensive, Easily deployable Cons: Improper PC hardware or network congestion can degrade video
quality and optimal control response times
6
Related Work Telemicroscopy over Internet2
Gemini Observatory NanoManipulator
Telescience Project – National Center for Microscopy and Imaging Research, UC San Diego
Ultrahigh Voltage Electron Microscope Research Center – Osaka University
Common Instrument Middleware Architecture (CIMA) – Indiana University
Tele-presence Microscopy – Argonne National Lab’s Advanced Analytical Electron Microscope facility
+) Novel applications for controlling instruments
+) All said “it works” over XYZ network paths and listed challenges they overcame
-) None have quantified performance in terms of network effects
-) None have considered user Quality of Experience (QoE)
Study Motivation: Understanding User and Network interplay can help us improve reliability and efficiency of Telemicroscopy and thus deliver optimum user QoE
7
→ user-activity (key strokes and mouse clicks) during a session involving n microscope functions→ average video image transfer rate at the microscope end→ network connection quality→ input-output scaling factor; unique to a microscope function→ seed image transfer rate; for quick screen refresh→ average video image transfer rate at the user end→ system-state control parameter dependent on user behavior; causes ± feedback in the control system
Telemicroscopy Session Model
(a) Session Model Parameters (b) Closed-loop Control System Representation
8
Telemicroscopy Session Model
(a) Session Model Parameters (b) Closed-loop Control System Representation
(c) Transfer Function
(d) End-user QoE relation in a Telemicroscopy session
Demand – Effort the user had to expend to perform n actions
Supply – Perceivable video image quality during the n actions
9
Telemicroscopy System States(Effects of H parameter)
(a) State Transitions
(b) System Supply-Demand Performance
10
Case Study: OSC Collaboration with OSU CAMM
OSU Center for Accelerated Maturation of Materials (CAMM) has acquired high-end Electron Microscopes Used for materials modeling studies at sub-angstrom level
OSC providing systems and networking support for Telemicroscopy OSCnet supporting end-to-end bandwidth requirements Image processing of samples (automation with MATLAB) for
Analytics service Telemicroscopy Demonstrations
Supercomputing, Tampa, FL (Nov 2006) Internet2 Fall Member Meeting, Chicago, IL (Dec 2006) Stark State University/Timken, Canton, OH (Mar 2007)
11
Telemicroscopy Testbed
Experiments to characterize session model parameters
Test cases with different network connections – CAMM requirements
(a) 1 Gbps LAN (Direct connection to Users in neighboring room) (b) Isolated LAN (Users in the same building ) (c) Public LAN (Users in different buildings on campus) (d) WAN (Users on the Internet)
Performance analysis goals Bandwidth, latency and packet loss levels for optimum user QoE Traffic characterization for studying inter-play between user control
(TCP traffic) and microscope response (UDP traffic)
12
WAN Testbed
(a) Setup
(b) WAN Path Performance
13
Performance Measurements Collected End-user QoE Measurements (Subjective Metrics)
Mean Opinion Scores (MOS) of “Novice” and “Expert” Users Time for completion of “basic” and “advanced” Tele-microscopy
tasks by Novice and Expert Users
Network Measurements (Objective Metrics) Collected using Ethereal/TCPdump and OSC ActiveMon
Metrics: Data rate, Protocols Summary
14
Network Connection Quality (ψnet) and User QoE (qmos)
qmos notably decreases with decrease in network connection quality User QoE is highly sensitive to network health fluctuations
Novice more liberal than Expert Time taken to complete a task increases with decrease in network connection
quality
NOTE: qmos of 5 corresponds to “at the microscope” QoE
15
Network Connection Quality (ψnet) and User Control (bin)
Mouse and Keyboard traffic is TCP traffic Higher TCP throughput on poor network connections
Increased user effort with keyboard and mouse on poor connections “Congestion begets more congestion”
Task-1 Task-2 Task-3
Task-1 Task-2 Task-3Task-1 Task-2 Task-3
1 Gbps LAN – Expert
Public 100 Mbps LAN – Expert100 Mbps WAN – Expert
User expends minimum effort with keyboard and
mouse to complete use-caseUser expends notably more
effort with keyboard and mouse to complete use-case
User expends a “lot” of effort with keyboard and
mouse to complete use-case
1400 B/s
140 s
900 B/s
100 s
60 B/s
60 s
16
Network Connection Quality (ψnet) and Image Transfer Rate (Δbout)
“At the microscope” QoE requires ~30 Mbps between user and microscope ends
Other WAN tests at SC06 (Tampa) and Internet2 FallMM (Chicago) to microscopes at CAMM (Columbus) Usable on ~(10-25) Mbps WAN connections Usable if one-way network delays within ~50ms; as much as ~20% UDP packet loss
tolerable if adequate bandwidth provisioned
17
OSC’s Remote Instrumentation Collabration Environment (RICE)
Leverages our user and network interplay studies for “reliable” and “efficient” Telemicroscopy sessions and thus delivers optimum user QoE
Customizable software on custom server-side hardware for Telemicroscopy Best of VNC and KVMoIP worlds
RICE Features Network-aware video encoding
Optimizes frame rates based on available network bandwidth Manual video-quality adjustment slider
Network-status and user-action blocking Warns user of network congestion that leads to unstable session state Blocks user-actions during extreme congestion scenarios and prevents
breakdown Collaboration tools
VoIP, Chat, Annotation, Command-abstraction Multi-user support
Control-lock passing, collaborators presence, colored-text chat conference Workflow and Image management
Simultaneously connects to multiple PCs, transfers images and transparently switches between them
18
RICE Demo Video
19
RICE use-cases for online learning
Remote students can view instructor (also remote!) controlling different types of scientific instruments Efficiently – with the appropriate video frames to match last mile network
capabilities Reliably – without worrying about damaging the instrument Multi-party VoIP and Chat collaboration Image Annotation
Instructor can pass control to students - train them to operate the instrument during the class
Students can conduct lab sessions at their assigned slots on the instruments
Students image files can be organized and hosted at a central server Analytics can be supported using a web-service to analyze the image data sets
20
Future Work Shared instrumentation uses OSC’s state-wide resources
Networking, Storage, HPC, Analytics
Cyberinfrastructure for Shared Instrumentation
21
Shared Instrumentation @ OSC
Plans underway to support shared instrumentation for - Ohio State University: CAMM Electron Microscopes, Chemistry
Department Spectrometers and Diffractometers, Astronomy Department Telescopes
Miami University: Electron Microscopes, EPR Spectrometers Ohio University: Nuclear Accelerator
22
Thank you for your attention!☺
Any Questions?