TY - GEN
T1 - Efficient Checkpoint under Unstable Power Supplies on NVM based Devices
AU - Liu, Jialin
AU - Sha, Edwin
AU - Zhuge, Qingfeng
AU - Xu, Rui
AU - Song, Yuhong
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - In smart cities, many embedded devices' power supply lacks stability, e.g., solar panel, since they are used outdoors. In these systems, power failures can cause data lost if the data is kept on volatile media like DRAM. Thus, using a backup mechanism to record the program status is necessary for the program going forward. Non-volatile memory (NVM) provides systems with a convenient method to persist the programs' status. However, NVM is notorious for poor write endurance. In this paper, we propose an automatic checkpoint mechanism that can reduce NVM write wear. Firstly, the checkpoint trigger code is instrumented into every basic block to achieve automatic runtime checkpoint activation. Then, when the program triggers checkpoint, the checkpoint procedure incrementally updates program state's backup to reduce redundant NVM writes. In this step, we select what to back up by determining the activated stack frame. Also, our mechanism has two backup copies to avoid power loss corrupting the backup content by interrupting the checkpoint procedure. Finally, we design an update log queue to fine-grained update of global data so that the NVM write can be reduced. Experiments show that our mechanism can effectively reduce total NVM write overhead with acceptable execution overhead.
AB - In smart cities, many embedded devices' power supply lacks stability, e.g., solar panel, since they are used outdoors. In these systems, power failures can cause data lost if the data is kept on volatile media like DRAM. Thus, using a backup mechanism to record the program status is necessary for the program going forward. Non-volatile memory (NVM) provides systems with a convenient method to persist the programs' status. However, NVM is notorious for poor write endurance. In this paper, we propose an automatic checkpoint mechanism that can reduce NVM write wear. Firstly, the checkpoint trigger code is instrumented into every basic block to achieve automatic runtime checkpoint activation. Then, when the program triggers checkpoint, the checkpoint procedure incrementally updates program state's backup to reduce redundant NVM writes. In this step, we select what to back up by determining the activated stack frame. Also, our mechanism has two backup copies to avoid power loss corrupting the backup content by interrupting the checkpoint procedure. Finally, we design an update log queue to fine-grained update of global data so that the NVM write can be reduced. Experiments show that our mechanism can effectively reduce total NVM write overhead with acceptable execution overhead.
KW - Backup
KW - Checkpoint
KW - Power Failure
KW - Stack
UR - https://www.scopus.com/pages/publications/85152225490
U2 - 10.1109/HPCC-DSS-SmartCity-DependSys57074.2022.00278
DO - 10.1109/HPCC-DSS-SmartCity-DependSys57074.2022.00278
M3 - 会议稿件
AN - SCOPUS:85152225490
T3 - Proceedings - 24th IEEE International Conference on High Performance Computing and Communications, 8th IEEE International Conference on Data Science and Systems, 20th IEEE International Conference on Smart City and 8th IEEE International Conference on Dependability in Sensor, Cloud and Big Data Systems and Application, HPCC/DSS/SmartCity/DependSys 2022
SP - 1846
EP - 1853
BT - Proceedings - 24th IEEE International Conference on High Performance Computing and Communications, 8th IEEE International Conference on Data Science and Systems, 20th IEEE International Conference on Smart City and 8th IEEE International Conference on Dependability in Sensor, Cloud and Big Data Systems and Application, HPCC/DSS/SmartCity/DependSys 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 24th IEEE International Conference on High Performance Computing and Communications, 8th IEEE International Conference on Data Science and Systems, 20th IEEE International Conference on Smart City and 8th IEEE International Conference on Dependability in Sensor, Cloud and Big Data Systems and Application, HPCC/DSS/SmartCity/DependSys 2022
Y2 - 18 December 2022 through 20 December 2022
ER -