Analytic modeling and optimization of the SSD performance in remote sensing systems

Abstract

The storage for mass data in remote sensing systems always restricts its development. Solid State drive (SSD) is an efficient storage method based on semiconductor device, with superior performance and flexible management. However, the performance of SSD is not stable as our expectations, especially when the user space fill to the full. Aiming at meeting the storage requirement of remote sensing data and making good use of SSD's advantage, the paper present an analytic modeling of SSD Performance in remote sensing systems, propose a optimization method to exploit the stable performance of SSD based on the model. We present the first precise write amplification modeling of SSD performance in consideration of various NAND Flash silicon semiconductor process, and validate under a COS SSD with SAR and optical images. In addition we also present a black-box models which predict performance degradation for both response time and throughput under simple different traffic conditions. Test results show the modeling can predict the inter-SSD operations and used to schedule the continuous IO. Based on the proposed model, the paper present an IO scheduling method to maintain a stable performance, especially write performance. Using trace-driven test in SSD black-box performance test system, we demonstrate how the proposed IO scheduling method can be parameterized to realize the sustainable performance of SSD.