Quality of service management

• US 8,250,197 B2
• Filed: 10/28/2008
• Issued: 08/21/2012
• Est. Priority Date: 10/28/2008
• Status: Active Grant

• Current Assignee: Vmware LLC (Broadcom, Inc.)
• Original Assignee: VMware, Inc. (Broadcom, Inc.)
• Inventors: Gulati, Ajay, Ahmad, Irfan
• Primary Examiner(s): Avellino, Joseph
• Assistant Examiner(s): Khan, Aftab Nasir
• Application Number:
  US12/260,041
• Publication Number:
  US 20100106816A1
• Time in Patent Office: 1,393 Days
• Field of Search: 718100-103, 711/167, 711/168, 711/151, 711/158
• US Class Current: 709/223
• CPC Class Codes: G06F 13/161 with latency improvement
• CPC Class Codes: G06F 2213/0064 Latency reduction in handling transfers
• CPC Class Codes: G06F 3/0659 Command handling arrangements, e.g. command buffers, queues, command scheduling
• CPC Class Codes: H04L 1/0018 based on latency requirement
• CPC Class Codes: H04L 47/56 implementing delay-aware scheduling
• CPC Class Codes: H04L 49/90 Buffering arrangements
• CPC Class Codes: H04L 49/901 using storage descriptor, e.g. read or write pointers
• CPC Class Codes: H04L 65/75 Media network packet handling
• CPC Class Codes: H04L 65/752 adapting media to network capabilities
• CPC Class Codes: H04L 65/80 Responding to QoS
• CPC Class Codes: H04L 67/025 for remote control or remote monitoring of applications
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
• CPC Class Codes: H04L 67/562 Brokering proxy services

Abstract

A method for providing quality of service to a plurality of hosts accessing a common resource is described. The common resource may be a middle-tier or back-end server. A client IO request is received at one host of the plurality of hosts from one of a plurality clients executing as software entities on respective hosts. The host determines whether an issue queue is full. The IO request is issued to the common resource when the issue queue is not full. A current average latency observed at the host and an adjusted window size is calculated, based at least in part on the current average latency. The issue queue is resized to correspond with the adjusted window size.

Claims

1. A method for providing quality of service to a plurality of hosts accessing a common resource, the common resource being a middle-tier or back-end server, the method comprising:

receiving from a client an IO request at one host of the plurality of hosts, the client being one client of a plurality of clients, the plurality of clients are virtual machines executing on the one host by way of virtualization software logically interposed and interfacing with the virtual machines and system hardware of the one host, each virtual machine having a guest operating system and at least an application;

determining whether an issue queue maintained by the one host is full, the issue queue maintaining dispatch and completion information about IO requests from the clients that have already been dispatched to and are pending at the common resource, the issue queue being considered full when a number of the IO requests for which dispatch and completion information is being maintained by the issue queue reaches a specified limit, the specified limit being a window size;

issuing the IO request to the common resource when the issue queue is not full the issue queue being individually used by the one host to determine when to issue the IO request from the one host;

adding an entry for the IO request to the issue queue upon issuing the IO request to the common resource, wherein entries in the issue queue include IO requests that have been dispatched but not completed, wherein completed IO requests are removed from the issue queue;

calculating a current average latency observed at the one host, the current average latency being an average of individual latencies observed in completing the IO requests; and

calculating an adjusted window size, the adjusted window size being based at least in part on the current average latency; and

setting the specified limit to correspond with the adjusted window size to control the number of IO requests that are added to the issue queue by the one host.

2. The method of claim 1, wherein the common resource is a data storage array.

3. The method of claim 1, wherein:

the common resource is a storage array; and

the IO request received from the client is a virtual IO request issued by the guest operating system executing in the virtual machine, the virtual IO request is a request for data from a virtual storage device, and the virtualization software maps the virtual storage device to the storage array by transforming the virtual IO request into a physical IO request that requests access from the storage array prior to issuing the IO request to the storage array.

4. The method of claim 1, wherein the current average latency at time t is (CAL_t) is calculated as follows:

CAL_t=(1−

α

)×

L+α

×

CAL<sub>t−

1</sub>

where L is a current latency value and α

is a constant smoothing parameter, and CAL<sub>t−

1</sub> is a previous value for the current average latency.

5. The method of claim 1, wherein the calculating of the adjusted window size further includes calculating a per-byte latency based on the current average latency, and basing the adjusted window size at least partially on the per-byte latency.

6. The method of claim 1, wherein the calculating of the adjusted window size includes factoring in an assigned share representing a relative level of priority of the one host.

7. The method of claim 1, wherein the control algorithm for the adjusted window size is computed according to the following formula:

8. The method of claim 1, wherein the calculating of the adjusted window size further includes limiting the adjusted window size to a maximum value.

9. The method of claim 1, wherein the adjusted window size is locally calculated at the one host computer.

10. The method of claim 1, wherein each host computer in the plurality of host computers calculates the adjusted window size independently.

11. A non-transitory computer-readable storage medium embodying program instructions for providing quality of service to a plurality of hosts accessing a common resource, the common resource being a middle-tier or back-end server, the program instructions causing the hosts to execute a method, the method comprising:

receiving from a client an IO request, the client being one client of a plurality of clients, the plurality of clients are virtual machines executing by way of virtualization software logically interposed and interfacing with the virtual machines and system hardware of the one host;

the program instructions are integrated with or operably linked to the virtualization software;

each virtual machine has a guest operating system and at least an application;

maintaining an issue queue containing dispatch and completion information about IO requests from the clients that have already been dispatch to and are pending at the common resource;

determining whether the issue queue is full, the issue queue being considered full when a number of the IO requests for which dispatch and completion information is being maintained by the issue queue reaches a specified limit, the specified limit being a window size;

issuing the IO request to the common resource only when the issue queue is not full, the issue queue being individually used by the one host to determine when to issue the IO request from the one host;

adding an entry for the IO request to the issue queue upon issuing the IO request to the common resource, wherein entries in the issue queue include IO requests that have been dispatched but not completed, wherein completed IO requests are removed from the issue queue;

calculating a current average latency observed at the one host, the current average latency being an average amount of time for a request issued to the common resource to be fulfilled by the common resource;

calculating an adjusted window size, the adjusted window size being based at least in part on the current average latency; and

setting the specified limit to correspond with the adjusted window size to control the number of IO requests that are added to the issue queue by the one host.

12. The non-transitory computer-readable storage medium of claim 11, wherein the common resource is a data storage array.

13. The non-transitory computer-readable storage medium of claim 11, wherein:

the common resource is a storage array; and

the IO request received from the client is a virtual IO request issued by the guest operating system executing in the virtual machine, the virtual IO request is a request for data from a virtual storage device, and the virtualization software maps the virtual storage device to the storage array by transforming the virtual IO request into a physical IO request that requests access from the storage array prior to issuing the IO request to the storage array.

14. The non-transitory computer-readable storage medium of claim 11, wherein the current average latency at time t is (CAL_t) is calculated as follows:

CAL_t=(1−

α

)×

L+α

×

CAL<sub>t−

1</sub>

where L is a current latency value and α

is a constant smoothing parameter, and CAL<sub>t−

1</sub> is a previous value for the current average latency.

15. The non-transitory computer-readable storage medium of claim 11, wherein the calculating of the adjusted window size further includes calculating a per-byte latency based on the current average latency, and basing the adjusted window size at least partially on the per-byte latency.

16. The non-transitory computer-readable storage medium of claim 11, wherein the calculating of the adjusted window size includes factoring in an assigned share representing a relative level of priority of the one host.

17. The non-transitory computer-readable storage medium of claim 11, wherein the control algorithm for the adjusted window size is computed according to the following formula:

18. The non-transitory computer-readable storage medium of claim 11, wherein the calculating of the adjusted window size further includes limiting the adjusted window size to a maximum value.