Virtual computing cluster resource scheduler
In some embodiments, a method for cluster resource scheduling, includes determining at least one load score; determining a memory score; determining an IO score; and monitoring a message bus for candidate messages when each of the at least one load score, memory score, and IO score is less than a pre-determined health threshold. In some embodiments, a host computer system for hosting a plurality of virtual machines (VMs), includes: a memory; a network adapter for communicating with the cluster by way of a message bus; a processor in electronic communication with the memory and the network adapter, wherein the processor is programmed to: determine at least one load score; determine a memory score; determine an IO score; and monitor the message bus for candidate messages when each of the at least one load score, memory score, and IO score is less than a pre-determined health threshold.
1. A computer-implemented method for cluster resource scheduling, comprising:
determining at least one load score (S L ), wherein the at least one load score S L is determined based on a measurement window (t) according to:
where L t is the load average value over measurement window t, N C is the number of processor cores, and α t is a pre-determined coefficient for the measurement window t;
determining a memory score (S M );
determining an input-output (IO) score (S IO );
monitoring a message bus for candidate messages when each of the at least one load score, memory score, and IO score is less than a pre-determined health threshold;
determining a duress status based on a metric in duress, wherein the metric in duress is determined by one or more of a load score, the memory score, and the IO score remains over the pre-determined health threshold for a duress period;
selecting a guest candidate for migration when a duress status is determined, wherein the guest candidate is selected by calculating a standard deviation (SD) of guest virtual machine (VM) values for the metric in duress, and selecting the guest candidate as a guest VM falling with an SD threshold; and
publishing a candidate message on the message bus, the candidate message comprising guest candidate information.
2. The computer-implemented method of claim 1 , wherein the load scores are determined for each of a 1, a 5, and a 15 minute measurement window.
3. The computer-implemented method of claim 2 , wherein the coefficients for each measurement window are: α 1 =1.5, α 5 =1.3, and α 15 =1.1.
4. The computer-implemented method of claim 1 , wherein the memory score (S M ) is based on a percentage of host memory in use.
5. The computer-implemented method of claim 1 , wherein the IO score (S IO ) is a CPU IO Wait Time percentage divided by 25.
6. The computer-implemented method of claim 1 , further comprising pausing VM creation activities when any one of the load score, the memory score, or the IO score is greater than the health threshold.
7. The computer-implemented method of claim 1 , further comprising ordering guest VMs based on guest VM values for the metric in duress.
8. The computer-implemented method of claim 1 , further comprising:
receiving a migration request from a target host of the cluster; and
migrating the guest candidate to the migration target.
9. The computer-implemented method of claim 1 , further comprising:
receiving a candidate message from a cluster host, the candidate message comprising guest candidate information;
determining a resource sufficiency based on the guest candidate information; and
sending a migration request to the cluster host indicating availability to receive a guest candidate.
10. A host computer system for hosting a plurality of virtual machines (VMs), the host computer system being a member of a cluster of hosts, the host computer system comprising:
a memory;
a network adapter for communicating with the cluster by way of a message bus;
a processor in electronic communication with the memory and the network adapter, wherein the processor is programmed to:
determine at least one load score (S L ), wherein the at least one load score S L is determined based on a measurement window (t) according to:
where L t is the load average value over measurement window t, N C is the number of processor cores, and α t is a pre-determined coefficient for the measurement window t;
determine a memory score (S M );
determine an input-output (IO) score (S IO );
monitor the message bus for candidate messages when each of the at least one load score, memory score, and IO score is less than a pre-determined health threshold;
determine a duress status based on a metric in duress, wherein the metric in duress is determined by one or more of a load score, the memory score, and the IO score remains over the pre-determined health threshold for a duress period;
select a guest candidate for migration when a duress status is determined, wherein the guest candidate is selected by calculating a standard deviation (SD) of guest VM values for the metric in duress, and selecting the guest candidate as a guest VM falling with an SD threshold; and
publish a candidate message on the message bus, the candidate message comprising guest candidate information.
11. The host computer system of claim 10 , wherein the load scores are determined for each of a 1, a 5, and a 15 minute measurement window.
12. The host computer system of claim 11 , wherein the coefficients for each measurement window are: α 1 =1.5, α 5 =1.3, and α 15 =1.1.
13. The host computer system of claim 10 , wherein the memory score (S M ) is based on a percentage of host memory in use.
14. The host computer system of claim 10 , wherein the IO score (S IO ) is a CPU IO Wait Time percentage divided by 25.