高可用性(HA)雲端計算平台為計算節點提供高可用性的保護，系統中至少包含兩個計算節點。每個節點通常涉及主機和虛擬機（ VM）保護。在我們之前的研究中， HA 系統能在建立 HA 叢集後提供主機和虛擬機器(VM)保護。 我們的 HA系統有 7 層保護：主機服務，主機操作系統，主機網絡，用於主機保護和 VM 網絡的主機電源， VM OS，用於 VM 保護的 VM 電源。 如果高可用性叢集中發生任何故障，系統將循序診斷來檢測故障。但是，在某些環境中，某些階層可能無法被系統保護。因此，在本研究中，我們提出了一種 HA 叢集的保護方法， HA 系統可以動態地改變保護範圍。 我們用了不同層級的保護在主機和 VM 上，來查找故障檢測時間和確定哪種組合提供最佳結果。 從最終結果中我們得知：當使用 2 個主機階層時，最佳的階層組合為主機服務-主機網路；而當使用 3 個主機階層時，最佳的階層組合為主機服務-主機網路-主機作業系統而當使用 4 個主機階層時，最佳的階層組合為主機服務-主機網路-主機作業系統-主机功率 。 此外，用戶能額外使用 1階層（ VM 網絡）來保護其 VM。該研究表明，系統可以透過有限的保護階層來檢測所有故障。因此，如果在故障發生在未受保護的階層，則系統可以防止最壞
情況發生。

A high availability (HA) cloud platform provides high availability protection to the compute pool, which involves at least two nodes. Each node typically involves host and virtual
machine (VM) protection. In our previous work, the HA system provides both host and VM protection once an HA cluster is created. Our HA system has 7 layer protections: host service,
host network, host OS, host power for host protection and VM network, VM OS, VM power for VM protection. If any failure occurred in the high availability cluster, the system uses
sequential diagnosis process to detect a failure. However, in some environment, some of the layer protections may not be supported. Therefore, in this study, we propose a way to modify
high availability protection in the HA cluster. So, the system can dynamically change the protection and the system can be more useful. We use various combination layer protections
for both hosts and VMs to find failure detection time and determine which combination layer protection that provides the best result. From the final result, we found out that the best
combination layer using 2 host layers is Host Service-Host Network; 3 host layers is Host Service-Host Network-Host OS; all layers (4 host layers) is Host Service-Host Network-Host
OS-Host Power. In addition, it could be better for the user to use 1 layer (VM network) to protect their VM. This study demonstrated that the system can detect all faults through limited layer protection. Thus, the system can prevent the worst case scenario if the failure occurred on the disabled layer protection.

[1] C. Y. Cheng, Z. J. Su, C. C. Chen, S. J. Chen, and W. J. Wang, “Supporting Software-Defined HA Clusters on Openstack Platform,” Proc. 2017 IEEE Int. Conf. Appl. Syst. Innov. Appl. Syst. Innov. Mod. Technol. ICASI 2017, pp. 268–271, 2017.
[2] Intel, “Intelligent Platform Management Interface (IPMI).” [Online]. Available: https://www.intel.com.tw/content/www/tw/zh/servers/ipmi/ipmi-technical-resources.html.
[3] V. R. Yerravalli and A. Tharigonda, “High Availability Cluster Failover Mechanism Using Artificial Neural Networks,” Proc. - 2015 IEEE Int. Conf. Cloud Comput. Emerg. Mark. CCEM 2015, pp. 81–84, 2016.
[4] L. G. Zhu, D. Z. Han, S. Z. Zhou, and C. S. Xie, “High Availability Cluster with Combining NAS and ISCSI,” Proc. 2006 Int. Conf. Mach. Learn. Cybern., vol. 2006, no. August, pp. 4455–4460, 2006.
[5] S. L. Campbell and R. Nikoukhah, “Auxiliary Signal Design for Failure Detection,” IEEE Control Syst., vol. 26, no. 3, pp. 124–126, 2006.
[6] D. Trihinas, “Monitoring Elastically Adaptive Cloud Services,” Master Comput. Sci. Thesis, vol. X, no. X, p. 74, 2014.
[7] Y. Wu, Y. Yuan, G. Yang, and W. Zheng, “An Adaptive Task-Level Fault-Tolerant Approach to Grid,” J. Supercomput., vol. 51, no. 2, pp. 97–114, 2010.
[8] Y. L. Lee, M. H. Ho, A. Suharsono, Y. C. Pan, W. J. Wang, and D. Liang, “NCU-HA: A Lightweight HA System for Kernel-Based Virtual Machine,” 2017 Int. Conf. Platf. Technol. Serv. PlatCon 2017 - Proc., 2017.
[9] C.D. Lu, "Scalable Diskless Checkpointing for Large Parallel Systems." [Online]. Available:
https://www.ideals.illinois.edu/bitstream/handle/2142/11054/Scalable%20Diskless%20Checkpointing%20for%20Large%20Parallel%20Systems.pdf?sequence=2&isAllowe
d=y