# US 2022/0035665 A1 · Sharing of compute resources between the virtualized radio access network (vran) and other workloads > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0035665 A1 - Document ID: US20220035665A1 - Status: Published application - Kind: A1 - Application number: 16/941,033 - Filed: 2020-07-28 - Published: 2022-02-03 - Earliest claimed: 2020-07-28 - Assignees: Microsoft Technology Licensing, LLC - Applicants: Microsoft Technology Licensing, LLC - Inventors: Xenofon FOUKAS; Bozidar RADUNOVIC - Web page: https://patentgrep.com/patent/US20220035665A1 ## CPC - G06F 9/4887: {involving deadlines, e.g. rate based, periodic} - G06F 9/505: {considering the load} - G06F 9/5077: {Logical partitioning of resources; Management or configuration of virtualized resources (specific details on emulation or internal functioning of virtual machines G06F9/455)} - G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound - G06N 20/00: Machine learning - G06N 20/20: Ensemble learning - G06F 2209/485: Resource constraint - G06F 2209/501: Performance criteria - G06F 2209/5019: Workload prediction - G06F 2209/5021: Priority - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract The present disclosure relates to systems and methods for sharing compute resources. The systems and methods may include identifying a plurality of workloads to complete by a deadline. The systems and methods may include generating a performance prediction for each workload of the plurality of workloads. The systems and methods may use the performance prediction to calculate a number of compute resources required for the plurality of workloads to complete by the deadline. The systems and methods may schedule the plurality of workloads across the number of compute resources. ## Claims ### Claim 1 A method, comprising: identifying a plurality of virtualized Radio Access Networks (vRAN) workloads and a plurality of other workloads; generating a performance prediction for each workload of the plurality of vRAN workloads that indicates a prediction for how long each workload of the plurality of vRAN workloads will take to complete; calculating based on the performance prediction a number of compute resources required for the plurality of vRAN workloads to use to complete by a deadline; scheduling the plurality of vRAN workloads across the number of compute resources required; determining a remaining number of available compute resources; and scheduling the plurality of other workloads across the remaining number of available compute resources. ### Claim 2 (depends on 1) The method of claim 1, wherein the plurality of vRAN workloads include one or more signal processing tasks for a base station or data link layer (L2) tasks. ### Claim 3 (depends on 2) The method of claim 2, wherein the deadline is a transmission deadline for transmission of the one or more signal processing tasks with the base station or reception of the one or more signal processing tasks with the base station. ### Claim 4 (depends on 2) The method of claim 2, wherein the performance prediction is generated by: using a machine learning model to generate a plurality of quantile decision trees, where each quantile decision tree is associated with an individual signal processing task; and identifying a tail latency of a leaf node in a corresponding quantile decision tree for the individual signal processing task for each of the one or more signal processing tasks. ### Claim 5 (depends on 4) The method of claim 4, wherein the tail latency indicates a predicted worst case execution time for each workload of the plurality of vRAN workloads. ### Claim 6 (depends on 4) The method of claim 4, wherein the machine learning model is trained with offline samples of signal processing tasks and online samples of signal processing tasks. ### Claim 7 (depends on 2) The method of claim 2, wherein calculating the number of compute resources required further comprises: determining a predicted base station worst case execution time for the base station by summing the performance prediction for each workload; and calculating the number of compute resources required to complete the plurality of vRAN workloads by the deadline based on the predicted base station worst case execution time. ### Claim 8 A server, comprising: a memory to store data and instructions; and at least one processor in communication with a plurality of compute resources and a virtualized Radio Access Network (vRAN), wherein the at least one processor is operable to: identify a plurality of vRAN workloads and a plurality of other workloads; generate a performance prediction for each workload of the plurality of vRAN workloads that indicates a prediction for how long each workload of the plurality of vRAN workloads will take to complete; calculate based on the performance prediction a number of compute resources of the plurality of compute resources required for the plurality of vRAN workloads to complete by a deadline; schedule the plurality of vRAN workloads across the number of compute resources required; determine a remaining number of available compute resources of the plurality of compute resources; and schedule the plurality of other workloads across the remaining number of available compute resources. ### Claim 9 (depends on 8) The server of claim 8, wherein the plurality of vRAN workloads include one or more signal processing tasks for a base station or data link layer (L2) tasks. ### Claim 10 (depends on 9) The server of claim 9, wherein the deadline is a transmission deadline for transmission of the one or more signal processing tasks with the base station or reception of the one or more signal processing tasks with the base station. ### Claim 11 (depends on 9) The server of claim 9, wherein the at least one processor is further operable to generate the performance prediction by: using a machine learning model to generate a plurality of quantile decision trees, where each quantile decision tree is associated with an individual signal processing task; and identifying a tail latency of a leaf node in a corresponding quantile decision tree for the individual signal processing task for each of the one or more signal processing tasks. ### Claim 12 (depends on 11) The server of claim 11, wherein the tail latency indicates a predicted worst case execution time for each workload of the plurality of vRAN workloads. ### Claim 13 (depends on 11) The server of claim 11, wherein the machine learning model is trained with offline samples of signal processing tasks and online samples of signal processing tasks. ### Claim 14 (depends on 9) The server of claim 9, wherein the at least one processor is further operable to calculate the number of compute resources required by: determining a predicted base station worst case execution time for the base station by summing the performance prediction for each workload; and calculating the number of compute resources required to complete the plurality of vRAN workloads by the deadline based on the predicted base station worst case execution time. ### Claim 15 A method, comprising: identifying a plurality of workloads to complete; generating a performance prediction for each workload of the plurality of workloads that indicates a prediction for how long each workload of the plurality of workloads will take to complete; identifying a quantile at a tail latency of the performance prediction for each workload of the plurality of workloads; calculating based on the quantile a number of compute resources required for the plurality of workloads to complete by a deadline; and scheduling the plurality of workloads across the number of compute resources. ### Claim 16 (depends on 15) The method of claim 15, wherein the tail latency is a range of values that indicate a slow runtime of the workload. ### Claim 17 (depends on 15) The method of claim 15, wherein the performance prediction is generated using a machine learning model. ### Claim 18 (depends on 17) The method of claim 17, wherein the machine learning model generates a plurality of quantile decision trees for each workload and the tail latency is identified in a leaf node of each quantile decision tree of the plurality of quantile decision trees. ### Claim 19 (depends on 15) The method of claim 15, wherein the plurality of compute resources includes one or more of a central processing unit (CPU) core, a field programmable gate array (FPGA), a graphics processing unit (GPU), or other hardware accelerators. ### Claim 20 (depends on 15) The method of claim 15, wherein the deadline is a transmission deadline for the plurality of workloads or a completion deadline for the plurality of workloads.