Sunday 02 February 2025
A team of researchers has made a significant breakthrough in the field of computing, developing a new scheduling algorithm that can reduce the number of energy-consuming copy instructions needed when multiple arrays are used in SIMD (Single Instruction, Multiple Data) processing.
Traditionally, computers use a single array to store data and perform calculations. However, as the demand for faster and more efficient computing continues to grow, researchers have turned to using multiple arrays to accelerate computations. This approach, known as multi-array processing, has several advantages, including increased parallelism and reduced memory access times.
However, there is a catch – when multiple arrays are used, data needs to be copied between them, which can consume a significant amount of energy. In fact, copy instructions can account for up to 70% of the total energy consumption in some cases.
To address this issue, the researchers have developed a new scheduling algorithm called MASIM (Multi-Array Scheduling for In-Memory Computing). This algorithm uses a combination of two priorities to determine which data should be copied and when. The first priority is based on the number of copy instructions needed, while the second priority takes into account the number of partners – or nodes that share fanouts with other nodes.
The researchers tested MASIM using several benchmarks, including complex mathematical calculations and neural network simulations. Their results show that MASIM can reduce the number of copy instructions by up to 63% compared to existing algorithms, leading to a significant reduction in energy consumption.
But how does it work? The algorithm starts by identifying which nodes are ready to be computed and which arrays they should be stored in. It then uses the two priorities mentioned earlier to determine which node and array combination is best. If there are multiple options with the same priority, the algorithm randomly selects one.
Once the node and array have been selected, the algorithm inserts the necessary copy instructions to move the data between arrays. This process is repeated until all nodes have been computed and stored in their respective arrays.
The researchers also developed an iterative improvement process that can further reduce the number of copy instructions by up to 11%. This process involves iteratively applying the scheduling algorithm to improve the results.
The implications of this research are significant, as it has the potential to significantly reduce the energy consumption of computers and other devices. In addition, the algorithm could be used in a variety of applications, including artificial intelligence, data processing, and scientific simulations.
Cite this article: “Energy-Efficient Multi-Array Scheduling Algorithm Reduces Copy Instructions by Up to 63%”, The Science Archive, 2025.
Computing, Scheduling Algorithm, Multi-Array Processing, Simd, Energy Consumption, Copy Instructions, Masim, Priority-Based, Node Selection, Iterative Improvement, Data Processing
