Jongouk Choi University of Central Florida NVMW23 The Problem IoT market is bottlenecked by batteries 2 Intermittently compute only when enough energy is secured in a capacitor ID: 1042842
Download Presentation The PPT/PDF document "CapOS : Capacitor Error Resilience for E..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. CapOS: Capacitor Error Resilience for Energy Harvesting SystemsJongouk Choi*, , *University of Central Florida, NVMW’23
2. The ProblemIoT market is bottlenecked by batteries2Intermittently compute only when enough energy is secured in a capacitor[NASA Mars Rovers ’20]https://mars.nasa.gov/mars2020/[SmartThings ’21]https://www.design-reuse.com/articles/47333/securing-smart-connected-homes-with-otp-nvm.html
3. KineticWiFi, RFSolarThermalWhat is Energy Harvesting System?3
4. Challenge of Energy Harvesting SystemsUnstable and Intermittent Power Input [Ma,HPCA’15] 4
5. Timeline of program execution: Power failure on which a processor loses all volatile dataProblem of Frequent Power Failure5HELLO,HELLO,WORL
6. What is Intermittent Computing?Non-VolatileMemory (NVM)Energy harvesting systemCapacitor (as energy buffer)Ambient energyCollect energy(device off)Run program(device on)6Intermittently compute only when enough energy is secured in a capacitor
7. Roll-Forward Recovery: Just-In-Time CheckpointVoltage MonitorBackup/recovery ControllerVolatile registerCoreTimeCapacitor ChargeVbkVoffNVMCheckpoint volatile data to NVM with secured energy7Checkpoint(){str R1 mem[100]str R2 mem[200]…}restore(){load mem[100] R1load mem[200] R2…}
8. Reliability IssuesDegradationEnvironmental factors(temperature, humidity, over-voltage, continuous charging/discharging, ripple current)[Teverovsky, TCPMT’14]8Reliability: Aging Issue of Capacitor
9. Capacitor ErrorVoltage MonitorBackup/recovery ControllerVolatile registerCoreTimeCapacitor ChargeVbkVoffNVMPower failure during checkpointingCheckpoint/Restore wrong data across power failure9Checkpoint(){str R1 mem[100]str R2 mem[200]…}restore(){load mem[100] R1load mem[200] R2…}Wrong recovery!!!
10. CapOS: Capacitor Error Resilience for Energy Harvesting Systems Capacitor error detection JIT checkpoint with ACK Execution mode changeForward Progress10
11. Challenge of Proactive Capacitor Error DetectionVoltage MonitorBackup/recovery ControllerVolatile registerCoreTimeCapacitor ChargeVbkVoffNVMEnergy in guardband cannot be used for forward progressFalse-positive errors can lead to poor performance11Secured higher voltageCheckpoint(){str R1 mem[100]str R2 mem[200]…}
12. Challenge of Reactive Error DetectionVoltage MonitorBackup/recovery ControllerVolatile registerCoreTimeCapacitor ChargeVbkVoffNVMCheckpoint/Restore wrong data across power failurePermanent data corruption12Checkpoint(){str R1 mem[100]str R2 mem[200]…}
13. Voltage MonitorBackup/recovery ControllerVolatile registerCoreTimeCapacitor ChargeVbkVoffNVMACK is corrupted when the capacitor error occurs.Protect all other volatile data13Checkpoint(){str R1 mem[100]str R2 mem[200]…str ACK mem[1000]}restore(){load mem[100] R1load mem[200] R2…load ACK mem[1000]}Safe Recovery!!Reactive Detection with ACK
14. CapOS: Capacitor Error Resilience for Energy Harvesting Systems Capacitor error detection JIT checkpoint with ACK Execution mode changeForward progress14Rollback or Roll-forward?
15. Timeline of program execution: Power failure on which a processor loses all volatile dataChallenge: Incorrect Recovery15HELLO,Wrong recovery!Checkpoint RegistersNon-VolatileMemory (NVM)Restore Registers
16. Timeline of program execution: Power failure on which a processor loses all volatile dataChallenge: Roll-Forward to Rollback Recovery?16HELLO,,,,,where to rollback?Checkpoint RegistersNon-VolatileMemory (NVM)Restore RegistersWrong Recovery: Region boundaryNo log?
17. Timeline of program execution: Power failure on which a processor loses all volatile dataCapOS: Boundary-Free Intermittent Computation17HELLO,,,,,Mode ChangeCheckpoint RegistersNon-VolatileMemory (NVM)Restore RegistersSafe RecoveryCheckpoint registers when the watchdog timer expiresLeverage the page protection mechanism (COW) with memory protection unitABCDRestore RegistersCheckpoint Registers
18. Evaluation SettingMSP430 (Target Board, Power Generator Board)We compared CapOS to Chinchilla (rollback recovery solution, OSDI’18) and Samoyed (roll-forward recovery solution, PLDI’19).18VinGnd
19. Performance Analysis (No Capacitor Error)19
20. 20Performance Analysis (Capacitor Error)
21. ConclusionEnergy harvesting systems can lose their volatile data due to the problem of capacitor degradation, which we call a capacitor error. To address the capacitor error, we introduce CapOS that can detect capacitor errors without causing wrong recovery problem, and ensure forward progress execution. 21