State-of-the-art storage devices that have parallel capability have significantly reduced the performance gap between processor and storage I/O. However, the internal parallelism makes it difficult to measure utilization that can be used as a basis of load balancing, which is a critical feature of performance improvement of parallel systems. When utilization of storage reaches to one hundred percent, the I/O saturation occurs, and then some of I/O loads need to be redirected to the other storage to improve the whole storage performance. There is, to my best knowledge, no studies in I/O saturation prediction of the flash SSDs regarding the internal parallelism that the previous HDD based measure cannot reflect. In this paper, we propose I/O saturation prediction method based on ANN (Artificial Neural Network) using kernel I/O statistics and various performance measures. We extracted I/O statistics and performance measures by conducting I/O workload simulation especially on NVMe (Non-Volatile Memory Express) flash SSD so that we can get as many observations from various I/O characteristics as flash SSD possibly can show. We constructed ANN model and compared with SVM (Support Vector Machine) model. The evaluation shows that ANN performs well compared to the existing utilization measure which assumes that HDD can only perform single instruction at a time.
|Title of host publication||2017 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2017|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||6|
|Publication status||Published - 2017 Nov 27|
|Event||2017 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2017 - Banff, Canada|
Duration: 2017 Oct 5 → 2017 Oct 8
|Name||2017 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2017|
|Other||2017 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2017|
|Period||17/10/5 → 17/10/8|
Bibliographical noteFunding Information:
This research was supported by the MSIT(Ministry of Science and ICT), Korea, under the SW Starlab support program(IITP-2017-0-00477) supervised by the IITP(Institute for Information & communications Technology Promotion).
© 2017 IEEE.
All Science Journal Classification (ASJC) codes
- Artificial Intelligence
- Computer Science Applications
- Human-Computer Interaction
- Control and Optimization