Automatic System Service Resource Management for Virtualizing Low-Latency Workloads that are Input/Output Intensive

• US 20180101486A1
• Filed: 12/15/2016
• Published: 04/12/2018
• Est. Priority Date: 10/06/2016
• Status: Active Grant

• Current Assignee: Vmware LLC (Broadcom, Inc.)
• Original Assignee: VMware, Inc. (Broadcom, Inc.)
• Inventors: LU, Xunjia, ZHENG, Haoqiang
• Granted Patent: US 10,452,572 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 13/161 with latency improvement
• CPC Class Codes: G06F 15/7875 for multiple contexts
• CPC Class Codes: G06F 2009/4557 Distribution of virtual machine instances; Migration and load balancing
• CPC Class Codes: G06F 9/45533 Hypervisors; Virtual machine monitors
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: G06F 9/4812 by interrupt, e.g. masked
• CPC Class Codes: G06F 9/5033 considering data affinity
• CPC Class Codes: G06F 9/5088 involving task migration

Abstract

An approach for providing low-latency and fast response times required by virtual machine (VM) telecommunication (telco) workloads is disclosed. In one embodiment, a kernel scheduler identifies hypervisor system contexts that are outside of traditional virtual central processing unit (CPU) contexts and servicing telco workloads. The scheduler grants each such context exclusive affinity to a respective physical CPU, up to a configurable maximum number of such grants. The scheduler also increases the CPU reservations in a system resource pool for the hypervisor by one CPU, indicating that there is one fewer physical CPU on which VMs may be placed or moved. The exclusive affinities and CPU reservations may be dynamically adjusted as, e.g., new latency-sensitive VMs are powered on with virtual CPUs that need to be granted exclusive affinity to physical CPUs or contexts no longer relevant to telco workloads are demoted to not have exclusive affinity to physical CPUs.

Claims

1. In a host computer having one or more physical central processing units (CPUs) that support execution of virtual computing instances, a method for reducing latency and jitter of an input/output (I/O) intensive workload, comprising:

identifying a first system context of a hypervisor, the first system context servicing the I/O intensive workload which runs in one of the virtual computing instances;

assigning the first system context to have exclusive affinity to one of the physical CPUs, wherein the assigning includes migrating running tasks and directing queued tasks and interrupt processing from the one of the physical CPUs to other physical CPUs.

2. The method of claim 1, wherein identifying the first system context that services the I/O intensive workload includes determining that a communication rate of the first system context with a virtual CPU is greater than a predefined threshold or determining that a CPU load associated with the first system context is greater than a predefined threshold.

3. The method of claim 1, wherein identifying the first system context that services the I/O intensive workload includes determining that the first system context is related to a second system context of the hypervisor whose communication rate with a virtual CPU is greater than a predefined threshold or whose associated CPU load associated is greater than a predefined threshold.

4. The method of claim 1, wherein identifying the first system context that services the I/O intensive workload includes:

receiving, from a driver, a notification of the first system context as being related to a virtual CPU; and

determining that a physical CPU load associated with the first system context is greater than a predefined threshold.

5. The method of claim 1, wherein the first system context is identified based on an associated virtual device or virtual device queue being tagged as involved in an I/O latency-sensitive workload.

6. The method of claim 1, further comprising,

increasing, by one physical CPU, a hypervisor resource pool allocation for the host computer, wherein the hypervisor resource pool is used by a management application to make virtual computing instance placement decisions.

7. The method of claim 1, wherein assigning the first system context exclusive affinity to the one of the physical CPUs places the first system context in a same last-level-cache (LLC) or Non-Uniform Memory Access (NUMA) domain as a virtual CPU to which the first system context communicates.

8. The method of claim 1, wherein:

a maximum number of system contexts that can be assigned to have exclusive affinity to the physical CPUs for a virtual computing instance is user-defined; and

the first system context is assigned to have exclusive affinity to the one of the physical CPUs responsive to determining that the maximum number of system contexts that can be assigned to have exclusive affinity to the physical CPUs has not been exceeded.

9. The method of claim 1, further comprising, responsive to determining that a new latency-sensitive virtual computing instances cannot be placed as a result of the first system context'"'"'s exclusive affinity to the one of the physical CPUs:

demoting the first system context to not have exclusive affinity to the one of the physical CPUs; and

assigning a virtual CPU of the new latency-sensitive virtual computing instance to have exclusive affinity to the one of the physical CPUs.

10. The method of claim 1, further comprising, responsive to identifying that the first system context no longer services an I/O intensive workload, demoting the first system context to not have exclusive affinity to the one of the physical CPUs.

11. The method of claim 1, wherein the first system context is one of a physical network interface card (NIC) context configured to process packets that are received or a virtual NIC context configured to process packets that are being transmitted.

12. A non-transitory computer-readable medium comprising instructions executable by a host computer, the host computer having one or more physical central processing units (CPUs) that support execution of virtual computing instances, wherein the instructions, when executed, cause the host computer to perform a method for reducing latency and jitter of an input/output (I/O) intensive workload, the method comprising:

identifying a first system context of a hypervisor, the first system context servicing the I/O intensive workload which runs in one of the virtual computing instances;

assigning the first system context to have exclusive affinity to one of the physical CPUs, wherein the assigning includes migrating running tasks and directing queued tasks and interrupt processing from the one of the physical CPUs to other physical CPUs.

13. The non-transitory computer-readable medium of claim 12, wherein identifying the first system context that services the I/O intensive workload includes one of:

determining that a communication rate of the system context with a virtual CPU is greater than a predefined threshold or determining that a CPU load associated with the first system context is greater than a predefined threshold;

or

determining that the first system context is related to a second system context of the hypervisor whose communication rate with the virtual CPU is greater than the predefined threshold or whose associated CPU load associated is greater than a predefined threshold.

14. The non-transitory computer-readable medium of claim 12, wherein identifying the first system context that services the I/O intensive workload includes:

receiving, from a driver, a notification of the first system context as being related to a virtual CPU; and

determining that a physical CPU load associated with the first system context is greater than a predefined threshold.

15. The non-transitory computer-readable medium of claim 12, wherein the first system context is identified based on an associated virtual device or virtual device queue being tagged as involved in an I/O latency-sensitive workload.

16. The non-transitory computer-readable medium of claim 12, the method further comprising, increasing, by one physical CPU, a hypervisor resource pool allocation for the host computer, wherein the hypervisor resource pool is used by a management application to make virtual computing instance placement decisions.

17. The non-transitory computer-readable medium of claim 12, wherein assigning the first system context exclusive affinity to the one of the physical CPUs places the first system context in a same last-level-cache (LLC) or Non-Uniform Memory Access (NUMA) domain as a virtual CPU to which the first system context communicates.

18. The non-transitory computer-readable medium of claim 12, the method further comprising, responsive to determining that a new latency-sensitive virtual computing instances cannot be placed as a result of the first system context'"'"'s exclusive affinity to the one of the physical CPUs:

demoting the first system context to not have exclusive affinity to the one of the physical CPUs; and

assigning a virtual CPU of the new latency-sensitive virtual computing instance to have exclusive affinity to the one of the physical CPUs.

19. The non-transitory computer-readable medium of claim 12, wherein the first system context is one of a physical network interface card (NIC) context configured to process packets that are received or a virtual NIC context configured to process packets that are being transmitted.

20. A host computer system, comprising:

one or more physical central processing units (CPUs) that support execution of virtual computing instances;

a memory, wherein the memory includes an application program configured to perform operations for reducing latency and jitter of an input/output (I/O) intensive workload, the operations comprising;

identifying a first system context of a hypervisor, the first system context servicing the I/O intensive workload which runs in one of the virtual computing instances, and

assigning the first system context to have exclusive affinity to one of the physical CPUs, wherein the assigning includes migrating running tasks and directing queued tasks and interrupt processing from the one of the physical CPUs to other physical CPUs.