Method and device for computing resource scheduling

• US 10,432,550 B2
• Filed: 07/07/2016
• Issued: 10/01/2019
• Est. Priority Date: 07/10/2015
• Status: Active Grant

• Current Assignee: Alibaba Group Holding Ltd.
• Original Assignee: Alibaba Group Holding Ltd.
• Inventors: Zhang, Yunyang, Yan, Huining, Li, Meng, Xu, Hongbin
• Primary Examiner(s): Tang, Karen C
• Application Number:
  US15/204,623
• Publication Number:
  US 20170012892A1
• Time in Patent Office: 1,181 Days
• Field of Search: 709231, 709224
• CPC Class Codes: G06F 2009/4557 Distribution of virtual machine instances; Migration and load balancing
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: G06F 9/5061 Partitioning or combining of resources
• CPC Class Codes: 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
• CPC Class Codes: H04L 47/70 Admission control; Resource allocation
• CPC Class Codes: H04L 47/785 among multiple network domains, e.g. multilateral agreements
• CPC Class Codes: H04L 47/823 Prediction of resource usage
• CPC Class Codes: H04L 47/83 based on usage prediction
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46
• CPC Class Codes: H04L 67/148 Migration or transfer of sessions

Abstract

A method, apparatus and device for scheduling resources of a cluster comprising a plurality of hosts, each running at least one instance, including acquiring a resource parameter of the cluster; calculating the number of predicted hosts in the cluster according to the resource parameter; determining to-be-migrated hosts and target hosts from the current hosts in the cluster when the number of current hosts in the cluster is greater than the number of predicted hosts; and migrating instances running on the to-be-migrated host to the target host.

Claims

1. A method for resource scheduling comprising:

acquiring, by a resource scheduling device, a resource configuration quantity of a cluster, the cluster including a plurality of hosts, the plurality of hosts including a plurality of current hosts each running one or more instances, the resource configuration quantity comprising a total capacity of each host in the cluster;

acquiring, by the resource scheduling device, an actual resource utilization quantity of each host in the cluster, the actual resource utilization quantity acquired by analyzing resources actively utilized by instances executing on each host;

generating, by the resource scheduling device, a resource parameter based on the resource configuration quantity and the actual resource utilization quantity;

calculating, by the resource scheduling device, a number of predicted hosts in the cluster according to the resource parameter by predicting the resource utilization quantities of instances in a given cluster, the predicting generating maximum value of resource utilization quantities of the hosts in a prospective time period based on the corresponding actual resource utilization quantities;

comparing, by the resource scheduling device, the predicted resource utilization quantities to resource configuration quantities of the plurality of hosts;

determining, by the resource scheduling device, one or more to-be-migrated hosts and one or more target hosts from the current hosts in the cluster when a number of current hosts in the cluster is greater than the number of predicted hosts; and

migrating, by the resource scheduling device, an instance of the one or more instances running on a to-be-migrated host of the one or more to-be-migrated hosts to a target host of the one or more target hosts when the number of current hosts is greater than the number of predicted hosts.

2. The method according to claim 1, wherein the calculating the number of predicted hosts in the cluster comprises:

performing a linear regression analysis on the actual resource utilization quantity of each of the one or more instances running on each of the plurality of current hosts to acquire a predicted resource utilization quantity of each of the one or more instances;

calculating, a sum of the predicted resource utilization quantities of the one or more instances running on each of the plurality of current hosts as a predicted resource utilization quantity of each of the plurality of current hosts;

calculating a sum of the predicted resource utilization quantities of the plurality of current hosts as a predicted resource utilization quantity of the cluster; and

calculating the number of predicted hosts in the cluster according to the resource configuration quantities of the plurality of hosts and the predicted resource utilization quantity of the cluster.

3. The method according to claim 2, wherein the determining the one or more to-be-migrated hosts and the one or more target hosts comprises:

calculating a difference between the number of current hosts and the number of predicted hosts as a number of to-be-migrated hosts in the cluster;

calculating a migration cost value for each of the plurality of current hosts; and

determining the to-be-migrated hosts in the cluster according to an order of the migration cost values of the plurality of current hosts and the calculated number of to-be-migrated hosts, and a number of remaining current hosts in the cluster as the one or more target hosts.

4. The method according to claim 3, wherein the calculating the migration cost value comprises:

acquiring a number of instances and a total storage space for each of the one or more instances running on each of the plurality of current hosts;

calculating a notification migration cost value for each of the plurality of current hosts according to the number of instances;

calculating a data migration cost value for each of the plurality of current hosts according to the total storage space for each of the one or more instances running; and

calculating a sum of the notification migration cost value and the data migration cost value as the migration cost value for each of the plurality of current hosts.

5. The method according to claim 2, further comprising:

sequencing, by the resource scheduling device, all the one or more instances running on the one or more to-be-migrated hosts according to a descending order of the predicted resource utilization quantities of the respective instances when the number of current hosts in the cluster is greater than the number of predicted hosts; and

sequentially calculating, by the resource scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, a corresponding target migration host of the one or more target hosts for each instance running on the one or more to-be-migrated hosts, the target migration host being a host of the one or more target hosts for which a predicted resource utilization quantity achieves a maximum improvement after the respective instance is migrated to the target migration host, and migrating each instance running on the one or more to-be-migrated hosts to the respective target migration host,

wherein the migrating the instance of the one or more instances running on the to-be-migrated host to the target host is migrating a first instance running on a first to-be-migrated host to the corresponding target migration host.

6. The method according to claim 1, further comprising:

controlling, by the resource scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, the to-be-migrated host of the one or more hosts to go offline from the cluster when the instance running on the to-be-migrated host is migrated to the target host.

7. The method according to claim 1, further comprising:

calculating, by the resource scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, an actual resource utilization rate corresponding to each of the one or more target hosts;

determining, by the resource scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, a first target host of the one or more target hosts for which an actual resource utilization rate exceeds a preset utilization rate threshold;

determining, by the resource scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, a first to-be-migrated instance from the one or more instances running on the first target host, such that the actual resource utilization rate of the first target host is determined to be smaller than the preset utilization rate threshold after the first to-be-migrated instance is migrated; and

migrating, by the resource scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, the first to-be-migrated instance to a second target host of the one or more target hosts, the second target host being determined to achieve a maximum improvement on the resource utilization rate after migrating the first to-be-migrated instance.

8. The method according to claim 1, further comprising:

determining, by the resource scheduling device when the number of current hosts is less than the number of predicted hosts, a first current host of the plurality of current hosts for which an actual resource utilization rate exceeds a preset utilization rate threshold;

determining, by the resource scheduling device when the number of current hosts is less than the number of predicted hosts, a group of instances from the one or more instances running on the first current host, such that the actual resource utilization rate of the first current host is determined to be smaller than the preset utilization rate threshold after the group of instances is migrated; and

adding, by the resource scheduling device, one or more new hosts from the cluster, when the number of current hosts is less than the number of predicted hosts, a number of the new hosts being equal to a difference between the number of predicted hosts and the number of current hosts, and migrating the group of instances to a second host of the one or more new hosts, the second host being determined such that an actual resource utilization rate of the second host does not exceed the preset utilization rate threshold after migration.

9. A scheduling device for resource scheduling on a cluster including a plurality of hosts, the device comprising:

one or more processors;

a network interface; and

a non-transitory memory storing computer-executable instructions executable by the processor, the instructions causing the device to;

acquire a resource configuration quantity of the cluster, the cluster including a plurality of hosts, the plurality of hosts including a plurality of current hosts each running one or more instances, the resource configuration quantity comprising a total capacity of each host in the cluster;

acquire an actual resource utilization quantity of each host in the cluster, the actual resource utilization quantity acquired by analyzing resources actively utilized by instances executing on each host;

generate a resource parameter based on the resource configuration quantity and the actual resource utilization quantity;

calculate a number of predicted hosts in the cluster according to the resource parameter by predicting the resource utilization quantities of instances in a given cluster, the predicting generating maximum value of resource utilization quantities of the hosts in a prospective time period based on the corresponding actual resource utilization quantities; and

comparing the predicted resource utilization quantities to resource configuration quantities of the plurality of hosts;

determine one or more to-be-migrated hosts and one or more target hosts from the current hosts in the cluster when a number of current hosts in the cluster is greater than the number of predicted hosts; and

migrate an instance of the one or more instances running on a to-be-migrated host of the one or more to-be-migrated hosts to a target host of the one or more target hosts when the number of current hosts is greater than the number of predicted hosts.

10. The scheduling device according to claim 9, wherein the acquiring instruction further causes the device to:

collect a resource configuration quantity of each of the plurality of hosts in the cluster; and

collect an actual resource utilization quantity of each of the one or more instances running on each of the plurality of current hosts in the cluster.

11. The scheduling device according to claim 10, wherein the calculating instruction further causes the device to:

perform a linear regression analysis on the actual resource utilization quantity of each of the one or more instances running on each of the plurality of current hosts to acquire a predicted resource utilization quantity of each of the one or more instances;

calculate a sum of the predicted resource utilization quantities of the one or more instances running on each of the plurality of current hosts as a predicted resource utilization quantity of each of the plurality of current hosts;

calculate a sum of the predicted resource utilization quantities of the plurality of current hosts as a predicted resource utilization quantity of the cluster; and

calculate the number of predicted hosts in the cluster according to the resource configuration quantities of the plurality of hosts and the predicted resource utilization quantity of the cluster.

12. The scheduling device according to claim 11, wherein the determining instruction further causes the device to:

calculate a difference between the number of current hosts and the number of predicted hosts as a number of to-be-migrated hosts in the cluster;

calculate a migration cost value for each of the plurality of current hosts; and

determine the to-be-migrated hosts in the cluster according to an order of the migration cost values of the plurality of current hosts and the calculated number of to-be-migrated hosts, and a number of remaining current hosts in the cluster as the one or more target hosts.

13. The scheduling device according to claim 12, wherein the calculating the migration cost value instruction further causes the device to:

acquire a number of instances and a total storage space for each of the one or more instances running on each of the plurality of current hosts;

calculate a notification migration cost value for each of the plurality of current hosts according to the number of instances;

calculate a data migration cost value for each of the plurality of current hosts according to the total storage space for each of the one or more instances running; and

calculate a sum of the notification migration cost value and the data migration cost value as the migration cost value for each of the plurality of current hosts.

14. The scheduling device according to claim 11, wherein the instructions further cause the device to:

sequence all the one or more instances running on the one or more to-be-migrated hosts according to a descending order of the predicted resource utilization quantities of the respective instances when the number of current hosts in the cluster is greater than the number of predicted hosts; and

sequentially calculate, by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, a corresponding target migration host of the one or more target hosts for each instance running on the one or more to-be-migrated hosts, the target migration host being a host of the one or more target hosts for which a predicted resource utilization quantity achieves a maximum improvement after the respective instance is migrated to the target migration host, and migrating each instance running on the one or more to-be-migrated hosts to the respective target migration host, the migrating the instance of the one or more instances running on the to-be-migrated host to the target host comprising migrating a first instance running on a first to-be-migrated host to the corresponding target migration host.

15. The scheduling device according to claim 9, wherein the instructions further cause the device to:

control, by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, the to-be-migrated host of the one or more hosts to go offline from the cluster when the instance running on the to-be-migrated host is migrated to the target host.

16. The scheduling device according to claim 9, wherein the instructions further cause the device to:

calculate, by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, an actual resource utilization rate corresponding to each of the one or more target hosts;

determine, by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, a first target host of the one or more target hosts for which an actual resource utilization rate exceeds a preset utilization rate threshold;

determine by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, a first to-be-migrated instance from the one or more instances running on the first target host, such that the actual resource utilization rate of the first target host is determined to be smaller than the preset utilization rate threshold after the first to-be-migrated instance is migrated; and

migrate, by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, the first to-be-migrated instance to a second target host of the one or more target hosts, the second target host being determined to achieve a maximum improvement on the resource utilization rate after migrating the first to-be-migrated instance.

17. The scheduling device according to claim 9, wherein the instructions further cause the device to:

determine, by the scheduling device when the number of current hosts is less than the number of predicted hosts, a first current host of the plurality of current hosts for which an actual resource utilization rate exceeds a preset utilization rate threshold;

determine, by the scheduling device when the number of current hosts is less than the number of predicted hosts, a group of instances from the one or more instances running on the first current host, such that the actual resource utilization rate of the first current host is determined to be smaller than the preset utilization rate threshold after the group of instances is migrated; and

add, by the scheduling device, one or more new hosts from the cluster, when the number of current hosts is less than the number of predicted hosts, a number of the new hosts being equal to a difference between the number of predicted hosts and the number of current hosts, and migrating the group of instances to a second host of the one or more new hosts, the second host being determined such that an actual resource utilization rate of the second host does not exceed the preset utilization rate threshold after migration.

18. A non-transitory computer-readable storage medium storing computer-executable instructions thereon that, when executed by a processor of a scheduling device, cause the scheduling device to perform resource scheduling on a cluster including a plurality of hosts, the scheduling comprising:

acquiring, by the scheduling device, a resource configuration quantity of the cluster, the cluster including a plurality of hosts, the plurality of hosts including a plurality of current hosts each running one or more instances, the resource configuration quantity comprising a total capacity of each host in the cluster;

acquiring, by the scheduling device, an actual resource utilization quantity of each host in the cluster, the actual resource utilization quantity acquired by analyzing resources actively utilized by instances executing on each host;

generating, by the scheduling device, a resource parameter based on the resource configuration quantity and the actual resource utilization quantity;

calculating, by the scheduling device, a number of predicted hosts in the cluster according to the resource parameter by predicting the resource utilization quantities of instances in a given cluster, the predicting generating maximum value of resource utilization quantities of the hosts in a prospective time period based on the corresponding actual resource utilization quantities;

comparing, by the scheduling device, the predicted resource utilization quantities to resource configuration quantities of the plurality of hosts;

determining, by the scheduling device, one or more to-be-migrated hosts and one or more target hosts from the current hosts in the cluster when a number of current hosts in the cluster is greater than the number of predicted hosts; and

migrating, by the scheduling device, an instance of the one or more instances running on a to-be-migrated host of the one or more to-be-migrated hosts to a target host of the one or more target hosts when the number of current hosts is greater than the number of predicted hosts.

19. The non-transitory computer-readable storage medium according to claim 18, the scheduling further comprising

controlling, by the scheduling device when the number of current hosts in the cluster is greater than the number of predicted hosts, the to-be-migrated host of the one or more hosts to go offline from the cluster when the instance running on the to-be-migrated host is migrated to the target host.