Granted Patent B2

US 12,566,629 · App. 17/972,570 · Granted Mar 3, 2026

Inventors: Chao Guo (Hong Kong, HK); Moshe Zukerman (Hong Kong, HK); Tianjiao Wang (Hong Kong, HK)

Assignee: City University of Hong Kong

G06F9/4881G06F9/5027

Abstract

Claims (41)

1 . A server comprising:

a plurality of processing modules each having predetermined resources for processing tasks handled by the server, wherein the plurality of processing modules are interconnected by communication links forming a network of processing modules having a Disaggregated Data Centers (DDC) architecture;

a scheduler module for scheduling and migrating tasks based on hardware information associated with the processing modules;

a monitor module for detecting topology information and load changes, including failure and repair of processing modules, and report the topology information and load changes to the scheduler module;

a central processing unit (CPU) and a memory storing instructions that when executed by the CPU, cause the CPU to:

by the scheduler module:

detect hardware information associated with the network of processing modules during an operation of the server;

analyze a resource allocation requests associated with a plurality of tasks and the hardware information; and

provide, based on the detection and analysis, a scheduler decision, wherein providing a scheduler decision includes for each task, allocating more than one processing modules in the network to handle the task based on resource availability of the plurality of processing modules and minimizing the blocking probability and number of task failing to complete their service while optimizing physical inter-resource traffic paths based on inter-resource traffic demand among different processing modules involved in handling the respective task of the plurality of tasks, wherein the blocking probability is one minus an acceptance ratio; and

based on changes to the hardware information, modifying the scheduler decision to:

for each task affected by the received change, reallocating more than one processing modules in the network to handle the task based on resource availability of the plurality of processing modules that have not experienced a failure or have been repaired and optimizing physical inter-resource traffic paths based on inter-resource traffic demand among different processing modules involved in handling the respective task of the plurality of tasks;

by the monitor module:

monitor for changes associated with the hardware information, wherein the hardware information includes topology of the DDC architecture, loading of each of the plurality of processing modules, and information related to failure and/or repairing of the network of processing modules; and

based on detected changes associated with the hardware information, notifying the scheduler module of the changes to perform a modification to the scheduler decision.

2 . The server in accordance with claim 1 , wherein the central processing unit (CPU) is further caused to analyze multiple resource allocation requests in a static scenario where the resource allocation requests are serviced in batches.

3 . The server in accordance with claim 2 , wherein the central processing unit (CPU) is further caused to analyze the multiple resource allocation requests based on a mixed-integer linear programming (MILP) method.

4 . The server in accordance with claim 3 , wherein the mixed-integer linear programming (MILP) method includes solving a MILP problem with varied weights in an objective function associated with a single-objective problem with a weighted sum, wherein the single-objective problem is converted from a multi-objective problem associated with multiple resource allocation requests in the static scenario.

5 . The server in accordance with claim 1 , wherein the centrals processing unit (CPU) is further caused to analyze multiple resource allocation requests in a dynamic scenario where the resource allocation requests sequentially and each resource allocation request is served over a predetermined period of time.

6 . The server in accordance with claim 5 , wherein the centrals processing unit (CPU) is further caused to schedule the resource allocation requests arriving in the dynamic scenario, based on the following conditions:

accepting the resource allocation request if sufficient resources are available upon arrival of the request; or

blocking the resource allocation request such that the request leaves the system without re-attempting;

wherein the resources are provided by a single processing module or a group of two or more processing modules involving inter-resource traffic.

7 . The server in accordance with claim 6 , wherein the centrals processing unit (CPU) is further caused to restore an accepted request being interrupted by hardware failure associated with resources allocated for handling the task, by excluding the processing module with hardware failure from the topology of the DDC architecture, and re-allocating resources for handling the accepted request.

8 . The server in accordance with claim 1 , wherein each of the plurality of processing modules includes plurality components of different types of resources, and wherein a single task including a request of more than one resource type is arranged to be processed by components of different types of resources in the plurality of processing modules in a disaggregated manner.

9 . A resource scheduling method for use in a server, wherein the server comprises a plurality of processing modules each having predetermined resources for processing tasks handled by the server, wherein the plurality of processing modules are interconnected by communication links forming a network of processing modules having a Disaggregated Data Centers (DDC) architecture; a scheduler module for scheduling and migrating tasks based on hardware information associated with the processing modules; and a monitor module for detecting topology information and load changes, including failure and repair of processing modules, and report the topology information and load changes to the scheduler module, the method comprising the steps of:

detecting, by the scheduler module, hardware information associated with the network of processing modules during an operation of the server;

analyzing, by the scheduler module, a resource allocation request a plurality of tasks and the hardware information; and

providing, by the scheduler module, based on the detection and analysis, a scheduler decision, wherein providing a scheduler decision include for each task, allocating more than one processing modules in the network to handle the task based on resource availability of the plurality of processing modules and minimizing the blocking probability and number of task failing to complete their service while optimizing physical inter-resource traffic paths based on inter-resource traffic demand among different processing modules involved in handling the respective task of the plurality of tasks, wherein the blocking probability is one minus an acceptance ratio;

subsequent to the providing, monitoring, by the monitoring module, for changes associated with the hardware information, wherein the hardware information includes topology of the DDC architecture, loading of each of the plurality of processing modules, and information related to failure and/or repairing of the network of processing modules;

based on detected changes associated with the hardware information, notifying, by the monitoring module, the scheduler module of the changes to perform a modification to the scheduler decision; and

based on changes to the hardware information, modifying the scheduler decision to: for each task affected by the received change, reallocating more than one processing modules in the network to handle the task based on resource availability of the plurality of processing modules that have not experienced a failure or have been repaired and optimizing physical inter-resource traffic paths based on inter-resource traffic demand among different processing modules involved in handling the respective task of the plurality of tasks.

10 . The resource scheduling method in accordance with claim 9 , wherein the step of analyzing the resource allocation request associated with each respective task and the hardware information comprising the step of analyzing multiple resource allocation requests in a static scenario where the resource allocation requests are serviced in batches.

11 . The resource scheduling method in accordance with claim 10 , wherein the multiple resource allocation requests are analyzed based on a mixed-integer linear programming (MILP) method.

12 . The resource scheduling method in accordance with claim 11 , wherein the mixed-integer linear programming (MILP) method includes solving a MILP problem with varied weights in an objective function associated with a single-objective problem with weighted sum, wherein the single-objective problem is converted from a multi-objective problem associated with multiple resource allocation requests in the static scenario.

13 . The resource scheduling method in accordance with claim 9 , wherein the step of analyzing the resource allocation request associated with each respective task and the hardware information comprising the step of analyzing multiple resource allocation requests in a dynamic scenario where the resource allocation requests are served sequentially and each resource allocation request is served over a predetermined period of time.

14 . The resource scheduling method in accordance with claim 13 , wherein the step of analyzing the resource allocation request associated with each respective task and the hardware information comprising the step of scheduling the resource allocation requests arriving in the dynamic scenario, based on the following conditions:

accepting the resource allocation request if sufficient resources are available upon arrival of the request; or

blocking the resource allocation request such that the request leaves the system without re-attempting;

wherein the resources are provided by a single processing module or a group of two or more processing modules involving inter-resource traffic.

15 . The resource scheduling method in accordance with claim 14 , wherein the step of analyzing the resource allocation request associated with each respective task and the hardware information comprising the step of restoring an accepted request being interrupted by hardware failure associated with resources allocated for handling the task, by excluding the processing module with hardware failure from the topology of the DDC architecture, and re-allocating resources for handling the accepted request.

16 . The resource scheduling method in accordance with claim 11 , wherein each of the plurality of processing modules includes plurality components of different types of resources, and wherein a single task including a request of more than one resource type is arranged to be processed by components of different types of resources in the plurality of processing modules in a disaggregated manner.

Assignments (1)

Continuity (1)

References Cited (69)