# US 2024/0201767 A1 · Utilizing device signals to improve the recovery of virtual machine host devices > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2024/0201767 A1 - Document ID: US20240201767A1 - Status: Published application - Kind: A1 - Application number: 18/084,822 - Filed: 2022-12-20 - Published: 2024-06-20 - Earliest claimed: 2022-12-20 - Assignees: Microsoft Technology Licensing, LLC - Applicants: Microsoft Technology Licensing, LLC - Inventors: Emma Sutherland BOYD; Shekhar AGRAWAL; Amruta Bhalchandra PATHAK; Yu YAO; Aravind Narayanan KRISHNAMOORTHY; Derek James BOYER; Binit Ranjan MISHRA; Gaurav JAGTIANI; Abhay Sudhir KETKAR; Tri Minh TRAN - Web page: https://patentgrep.com/patent/US20240201767A1 ## CPC - G06F 1/30: Means for acting in the event of power-supply failure or interruption, e.g. power-supply fluctuations (for resetting only G06F1/24) - G06F 9/45533: {Hypervisors; Virtual machine monitors} - G06F 11/0721: {within a central processing unit [CPU]} - G06F 11/0793: {Remedial or corrective actions (recovery from an exception in an instruction pipeline G06F9/3861; by retry G06F11/1402; for recovering from a failure of a protocol instance or entity H04L69/40)} - G06F 11/1438: {Restarting or rejuvenating} - G06F 11/1441: {Resetting or repowering} - G06F 11/3006: {where the computing system is distributed, e.g. networked systems, clusters, multiprocessor systems (multiprogramming arrangements G06F9/46; allocation of resources G06F9/50)} - G06F 11/3062: {where the monitored property is the power consumption (power management in a computing system G06F1/3203)} - G06F 11/301: {where the computing system is a virtual computing platform, e.g. logically partitioned systems (virtual machines G06F9/45533; logical partitioning of resources G06F9/5077)} - G06F 2201/81: Threshold - G06F 2201/815: Virtual - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract The present disclosure relates to utilizing a host failure recovery system to efficiently and accurately determine the health of host devices. For example, the host failure recovery system detects when a host server is failing by utilizing a power failure detection model that determines whether a host server is operating in a healthy power state or an unhealthy power state. In particular, the host failure recovery system utilizes a multi-layer power failure detection model that determines power-draw failure events on a host device. The failure detection model determines, with high confidence, the health of a host device based on power-draw signals and/or usage characteristics of the host device. Additionally, the host failure recovery system can initiate a quick recovery of a failing host device. ## Claims ### Claim 1 A computer-implemented method, including: receiving, at a management entity, a power-draw signal corresponding to a power draw by a host system on a host device; determining, at the management entity, that the host device is unhealthy using a power failure detection model that determines power-draw failings of host devices based on power-draw signals; and initiating, from the management entity, a recovery process to recover the host device based on determining that the host device is unhealthy. ### Claim 2 (depends on 1) The computer-implemented method of claim 1, wherein the power failure detection model includes a first rule-based layer and a second machine-learning layer, and the computer-implemented method further including determining whether to utilize the first rule-based layer or the second machine-learning layer to determine that the host device is unhealthy based on usage characteristics and the power-draw signal received from the host device. ### Claim 3 (depends on 2) The computer-implemented method of claim 2, further including: determining to utilize the first rule-based layer based on identifying a rule that corresponds to the power-draw signal from the host device; and determining that the host device is unhealthy based on the rule being satisfied by the power-draw signal. ### Claim 4 (depends on 2) The computer-implemented method of claim 2, further including: determining to utilize the second machine-learning layer having a power failure machine-learning model based on failing to identify a rule that corresponds to the power-draw signal or usage characteristics of the host device; determining a power-draw range for the host device using the power failure machine-learning model for the host device based on the usage characteristics of the host device; and determining that the host device is unhealthy based on the power-draw signal being outside the power-draw range for the host device. ### Claim 5 (depends on 4) The computer-implemented method of claim 4, wherein the power-draw signal and the usage characteristics received at the management entity correspond to operations being executed by a host processor of the host system of the host device. ### Claim 6 (depends on 2) The computer-implemented method of claim 2, wherein the usage characteristics of the host device include hardware information of the host device, software information of the host device, and workload information of virtual machines running on the host device. ### Claim 7 (depends on 1) The computer-implemented method of claim 1, wherein: the power-draw signal of the host system is received at the management entity from an auxiliary service controller on the host device; and the host device includes a separate processor and power supply for the auxiliary service controller from the host system. ### Claim 8 (depends on 7) The computer-implemented method of claim 7, wherein the auxiliary service controller determines the power-draw signal utilizing a power measurement device on the host device. ### Claim 9 (depends on 1) The computer-implemented method of claim 1, further including providing recovery instructions to the host system via an auxiliary service controller on the host device. ### Claim 10 (depends on 1) The computer-implemented method of claim 1, wherein initiating the recovery process includes providing, from the management entity, recovery instructions to the host device to reboot the host system. ### Claim 11 (depends on 1) The computer-implemented method of claim 1, further including: receiving, at the management entity, a second power-draw signal corresponding to a second host system on a second host device; determining, at the management entity, that the second host device is healthy by utilizing the power failure detection model based on the second power-draw signal; and delaying the recovery process to recover the second host device based on determining that the second host device is healthy. ### Claim 12 A system including: at least one processor at a server device; and a computer memory including instructions that, when executed by the at least one processor at the server device, cause the system to carry out operations including: receiving, at a management entity, a power-draw signal corresponding to a power draw by a host system on a host device; determining, at the management entity, that the host device is unhealthy using a power failure detection model that determines power-draw failings of host devices based on power-draw signals or usage characteristics of the host devices; and initiating, from the management entity, a recovery process to recover the host device based on determining that the host device is unhealthy. ### Claim 13 (depends on 12) The system of claim 12, wherein the power failure detection model determines that the host device is unhealthy based on the power-draw signal of the host device being below a lower power-draw threshold determined by a power failure machine-learning model for the host device. ### Claim 14 (depends on 12) The system of claim 12, wherein workload information from the usage characteristics of the host device includes a number of operations on the host device and operation types on the host device. ### Claim 15 (depends on 12) The system of claim 12, further including instructions that, when executed by the at least one processor, cause the system to carry out additional operations including delaying initiating the recovery process based on detecting a mitigation event corresponding to the host device. ### Claim 16 (depends on 12) The system of claim 12, wherein: the host device and the management entity are located in a data center; and the management entity manages multiple host devices, and the host device is one of the multiple host devices. ### Claim 17 (depends on 12) The system of claim 12, further including instructions that, when executed by the at least one processor, cause the system to carry out additional operations including: determining not to utilize a first rule-based layer of the power failure detection model; and utilizing a second machine-learning layer of the power failure detection model to determine that the host device is unhealthy in response to determining not to utilize the first rule-based layer of the power failure detection model. ### Claim 18 A computer-implemented method, including: receiving, at a management entity, a power-draw signal corresponding to a power draw by a host processor and host memory on a host device; generating a power failure detection model that includes a first rule-based layer and a second machine-learning layer having a power failure machine-learning model, wherein the power failure detection model determines power-draw failings of host devices based on power-draw signals and usage characteristics of the host devices; determining, at the management entity, that the host device is unhealthy by comparing the power-draw signal to a power-draw range determined by the power failure detection model for the host device based on one or more usage characteristics of the host device; and initiating, from the management entity, a recovery process to recover the host device based on determining that the host device is unhealthy. ### Claim 19 (depends on 18) The computer-implemented method of claim 18, wherein the power failure machine-learning model determines that the host device is unhealthy based on the power-draw signal of the host device being below a lower threshold of the power-draw range determined by the power failure machine-learning model for the host device. ### Claim 20 (depends on 18) The computer-implemented method of claim 18, wherein the power failure machine-learning model determines that the host device is unhealthy based on the power-draw signal of the host device being above an upper threshold of the power-draw range determined by the power failure machine-learning model for the host device.