Mementos System Support for Long-Running Computations on RFID-Scale - PowerPoint PPT Presentation

Mementos System Support for Long-Running Computations on RFID-Scale Devices pmn R Benjamin Ransford*, UMass Amherst ä~Äo ê~ío êó Jacob Sorber, Dartmouth College Kevin Fu, UMass Amherst http://spqr.cs.umass.edu/mementos ASPLOS XVI — March 8, 2011 Any opinions, findings, and conclusions expressed in this material are those CNS-0627529, CNS-0845874, CNS-0923313, Grad Res. Fel. of the authors and do not necessarily reflect the views of the NSF.

Ubiquitous Computing “... the most powerful things are those that are effectively invisible in use.” PARC — Mark Weiser (PARC, 1988) Problem: Batteryless invisible computer ⇒ constant reboots 2 Mementos — Ransford — ASPLOS XVI

Baby Steps Toward Ubicomp 1. Take Emerging Platform 3 Mementos — Ransford — ASPLOS XVI

Baby Steps Toward Ubicomp 1. Take Emerging Platform 2. Add Robustness Mechanism 3 Mementos — Ransford — ASPLOS XVI

Baby Steps Toward Ubicomp 1. Take Emerging Platform 2. Add Robustness Mechanism 3. Provide Simulation Tools 3 Mementos — Ransford — ASPLOS XVI

1 . E m e RFID-Scale Devices r g i n g P l a t f o r m Radio (RF) harvester x 0 1 d e i f i n g a m Energy buffer Reprogrammable (capacitor) microcontroller (~1 MHz) w/ on-chip flash Moo WISP: Hong Zhang 4 Mementos — Ransford — ASPLOS XVI

1 . E m e RFID-Scale Devices r g i n g P l a t f o r m Radio (RF) harvester x 0 1 d e i f i n g a m Energy buffer Reprogrammable (capacitor) microcontroller (~1 MHz) w/ on-chip flash Fills quickly, low capacity Moo WISP: Hong Zhang 4 Mementos — Ransford — ASPLOS XVI

1 . E m e RFID-Scale Devices r g i n g P l a t f o r m Radio (RF) harvester x 0 1 d e i f i n g a m Energy buffer Reprogrammable (capacitor) microcontroller (~1 MHz) w/ on-chip flash Fills quickly, Frequent reboots low capacity Moo WISP: Hong Zhang 4 Mementos — Ransford — ASPLOS XVI

ball: clipart.pierceinternet.com 5 Mementos — Ransford — ASPLOS XVI

REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOTREBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT REBOOT ball: clipart.pierceinternet.com 5 Mementos — Ransford — ASPLOS XVI

Robustness Under RF Harvesting • Typical approach: constrain the problem • Mementos: relax constraints to make general-purpose computation feasible 300 ms 6 Mementos — Ransford — ASPLOS XVI

Unpredictable Energy Morass Voltage Infinite energy; constant voltage Time 7 Mementos — Ransford — ASPLOS XVI

Unpredictable Energy Morass Voltage Infinite energy; constant voltage Time vs. Voltage (40 seconds) Time 7 Mementos — Ransford — ASPLOS XVI

2 . R o b u s t n Mementos Approach e s s M e c h a n i s m 3.6 3 2.5 Voltage (V) 2 1.5 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) • Checkpoint when failure appears imminent • Spread computation across reboots Movie poster: publispain.com 8 Mementos — Ransford — ASPLOS XVI

2 . R o b u s t n Mementos Approach e s s M e c h a n i s m 3.6 3 2.5 Voltage (V) 2 1.5 1 Checkpoint region 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) • Checkpoint when failure appears imminent • Spread computation across reboots Movie poster: publispain.com 8 Mementos — Ransford — ASPLOS XVI

2 . R o b u s t n Mementos Approach e s s M e c h a n i s m 3.6 3 2.5 Voltage (V) Checkpoint C 2 1.5 1 Checkpoint region 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) • Checkpoint when failure appears imminent • Spread computation across reboots Movie poster: publispain.com 8 Mementos — Ransford — ASPLOS XVI

2 . R o b u s t n Mementos Approach e s s M e c h a n i s m 3.6 ... 3 2.5 Voltage (V) Checkpoint C 2 1.5 1 Checkpoint region 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) • Checkpoint when failure appears imminent • Spread computation across reboots Movie poster: publispain.com 8 Mementos — Ransford — ASPLOS XVI

2 . R o b u s t n Mementos Approach e s s M e c h a n i s m Restore 3.6 ... 3 2.5 Voltage (V) Checkpoint C 2 1.5 1 Checkpoint region 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) • Checkpoint when failure appears imminent • Spread computation across reboots Movie poster: publispain.com 8 Mementos — Ransford — ASPLOS XVI

Running Example: CRC • Compute CRC16-CCITT checksum over 2 KB data • Tight nested loops • 575,000 CPU cycles ~ 575 ms 9 Mementos — Ransford — ASPLOS XVI

Running Example: CRC • Compute CRC16-CCITT checksum over 2 KB data • Tight nested loops Reboots every O(100) ms! • 575,000 CPU cycles ~ 575 ms 9 Mementos — Ransford — ASPLOS XVI

How to Use Mementos Programmer Mementos (our contributions) Write C code Choose params 10 Mementos — Ransford — ASPLOS XVI

How to Use Mementos Programmer Mementos (our contributions) Write Instrument w/ energy checks C code (via LLVM passes) Choose Simulate program params 10 Mementos — Ransford — ASPLOS XVI

How to Use Mementos Programmer Mementos (our contributions) Write Instrument w/ energy checks C code (via LLVM passes) Choose Simulate program params Suggest params 10 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (1/2) Programmer 1) Instrumentation strategy Write C code Choose params 11 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (1/2) Programmer 1) Instrumentation strategy Write C code Choose params Checkpoint? 11 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (1/2) Programmer 1) Instrumentation strategy Write C code Choose params Checkpoint? Checkpoint? 11 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write C code Choose params 12 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write C code 3.6 3 2.5 Voltage (V) Choose 2 DEATH 1.5 params 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) 12 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write CKPT C code 3.6 3 2.5 Voltage (V) Choose 2 DEATH 1.5 params 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) 12 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write C code Wasted! 3.6 { 3 CKPT 2.5 Voltage (V) Choose 2 DEATH 1.5 params 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) 12 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write C code 3.6 3 CKPT 2.5 Voltage (V) Choose 2 DEATH 1.5 params 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) 12 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write C code 3.6 3 2.5 Voltage (V) Choose 2 CKPT DEATH 1.5 params 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) 12 Mementos — Ransford — ASPLOS XVI

Choosing Parameters (2/2) Programmer 2) Checkpoint threshold V thresh Write C code 3.6 3 2.5 Voltage (V) CKPT Choose 2 DEATH 1.5 params 1 0.5 0 3155 3170 3180 3190 3200 3210 3220 Time (ms) 12 Mementos — Ransford — ASPLOS XVI

Mementos System Support for Long-Running Computations on RFID-Scale - PowerPoint PPT Presentation

Mementos System Support for Long-Running Computations on RFID-Scale Devices pmn R Benjamin Ransford*, UMass Amherst ~o ~o Jacob Sorber, Dartmouth College Kevin Fu, UMass Amherst http://spqr.cs.umass.edu/mementos ASPLOS XVI

TECHNOLOGY AT MASSEY UNIVERSITY A UNIQUE EXPERIENCE 1965-1995 Personal memories and mementos of

Key Management and Distribution Class 5 Stallings: Ch 14 1 CIS-5373: 10.February.2020

Chapter 7 by David G. Messerschmitt Understanding Networked Applications: A First Course

Cryptography and Network Distribution Security No Singhalese, whether man or woman, would Chapter

H P N H P N H P N H P N CERN igh igh erformance

Automatic Verification of Finite State Concurrent Systems Edmund M. Clarke, Jr. Computer Science

DECnet-Plus und TCP/IP Evolution of DNA DNA Phases Phase I DECUS IT - Symposium - 1976

56 C H A P T E R Chapter Goals Discuss the SNMP Management Information Base. Describe SNMP

Frame Relay Bigger, Longer, Uncut 2005/03/11 (C) Herbert Haas What is Frame Relay?

Shape-Based Quality Metrics for Large Graph Visualization* Peter Eades 1 Seok-Hee Hong 1 Karsten

System and Network Management Network Management : ability to monitor, control and plan the

XML and Databases Chapter 2: XML II: Entities and Marked Sections Prof. Dr. Stefan Brass

* Kurose and Ross, Computer Networking

Closing open SDL-systems for model checking with DTSpin Natalia Ioustinova Natalia Sidorova

Inside PDF Lecture @21C3 The Portable Document Format A Short Introduction Maik Musall

University of Manchester CS3282: Digital Communications 06 Section 9: Multi-level digital

M ARIA : Modular Reachability Analyser for Algebraic System Nets Marko Mkel Laboratory for

7. Introduction to image compression Image data is fundamentally different from text. Scale

SNMP Simple Network Management Protocol Computer Center, CS, NCTU Network Management The

Community-Driven Adaptation: Automatic Content Adaptation in Pervasive Environments Iqbal

FOCUS What are your hobbies or what do you love to do? What brings you excitement? What is your

USING QUALITY CONTROL CHARTS TO SEGMENT ROAD SURFACE CONDITION DATA Amin El Gendy D Doctoral

Finiteness Conditions on the Ext Algebra of a Monomial Algebra Ellen Kirkman kirkman@wfu.edu

Experiments in High-Frequency Trading: Testing the Frequent Batch Auction Eric M. Aldrich 1 opez