TY - GEN
T1 - Towards Scalable Quantum Circuit Simulation via RDMA
AU - Hsu, Chia Hsin
AU - Wang, Chuan Chi
AU - Hsu, Nai Wei
AU - Tu, Chia Heng
AU - Hung, Shih Hao
N1 - Publisher Copyright:
© 2023 ACM.
PY - 2023/8/6
Y1 - 2023/8/6
N2 - The trend towards optimization of quantum circuit simulators (QCS) has been driven by the rapid development of quantum computing, as there is a growing interest in creating high-efficiency QCS for quantum computing systems and applications. Despite this growth, the existing classical simulators pose challenges in simulating large quantum operations due to the exponential growth of memory and computation resources required. To overcome these issues, some studies have explored the use of non-volatile memory, such as solid-state disks (SSDs), to store a large amount of state vector while maintaining cost efficiency. However, such highly frequent read-and-write operations in large quantum simulations can quickly deplete the lifespan of SSDs, leading to overestimated practical efficiency and environmental concerns. Alternatively, Remote Direct Memory Access (RDMA) can be used to expand memory capacity, allowing computers in a network to access remote memories with low latency and high bandwidth. Therefore, we propose using one-sided RDMA operations along with a series of optimizations to expand memory capacity and efficiently maximize the utilization of unused memory assets. The experimental results demonstrate that our approaches can be practically scaled up without any lifespan consumption. The proposed optimization can get a 3.0x speedup compared to the local memory approach and a maximum speedup of 2.0x compared to the naive RDMA-based approach, as observed in our benchmark test.
AB - The trend towards optimization of quantum circuit simulators (QCS) has been driven by the rapid development of quantum computing, as there is a growing interest in creating high-efficiency QCS for quantum computing systems and applications. Despite this growth, the existing classical simulators pose challenges in simulating large quantum operations due to the exponential growth of memory and computation resources required. To overcome these issues, some studies have explored the use of non-volatile memory, such as solid-state disks (SSDs), to store a large amount of state vector while maintaining cost efficiency. However, such highly frequent read-and-write operations in large quantum simulations can quickly deplete the lifespan of SSDs, leading to overestimated practical efficiency and environmental concerns. Alternatively, Remote Direct Memory Access (RDMA) can be used to expand memory capacity, allowing computers in a network to access remote memories with low latency and high bandwidth. Therefore, we propose using one-sided RDMA operations along with a series of optimizations to expand memory capacity and efficiently maximize the utilization of unused memory assets. The experimental results demonstrate that our approaches can be practically scaled up without any lifespan consumption. The proposed optimization can get a 3.0x speedup compared to the local memory approach and a maximum speedup of 2.0x compared to the naive RDMA-based approach, as observed in our benchmark test.
UR - http://www.scopus.com/inward/record.url?scp=85174272749&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85174272749&partnerID=8YFLogxK
U2 - 10.1145/3599957.3606215
DO - 10.1145/3599957.3606215
M3 - Conference contribution
AN - SCOPUS:85174272749
T3 - 2023 Research in Adaptive and Convergent Systems RACS 2023
BT - 2023 Research in Adaptive and Convergent Systems RACS 2023
PB - Association for Computing Machinery, Inc
T2 - 2023 Research in Adaptive and Convergent Systems, RACS 2023
Y2 - 6 August 2023 through 10 August 2023
ER -