Dynamic allocation of queue depths for virtual functions in a converged infrastructure

1. An information handling system comprising:

a processor;

a memory system comprising an operating system (OS) and at least one device driver;

a plurality of host devices communicating input/output (I/O) requests with a plurality of virtual functions, respectively, wherein each host device comprises a corresponding device driver;

a peripheral component interconnect express (PCIe) storage controller communicatively coupled to the plurality of host devices and comprising:

the plurality of virtual functions; and

a controller firmware comprising: a command monitoring module; a queue management module; and a queue depth profiler;

wherein the (PCIe) storage controller executes code of the controller firmware, which configures the (PCIe) storage controller to:

determine for each virtual function (VF), using the command monitoring module, a value for at least one parameter associated with unfulfilled, outstanding requests, the at least one parameter including a count of total outstanding requests;

compare parameter values associated with outstanding requests across the plurality of virtual functions;

determine a relative demand for request queue depth for each respective VF based on the compared parameter values;

determine whether a threshold change in relative demand for at least one VF is detected; and

in response to detecting the threshold change in relative demand for at least one VF:

calculate for each VF, based on the relative demand, desired queue depth values, each representing a maximum number of outstanding commands that can be allocated to a respective VF;

initiate a “quiesce I/O” mode and initiate completion of pending requests for a first processing state; and

in response to successful completion of the pending requests for the first processing state: updates queue depth values from previous queue depth values to the desired queue depth values; resumes I/O processing queues; and initiates a second processing state to continue processing requests using current, desired queue depth values.

2. The IHS of claim 1 , further comprising a Small Computer System Interface (SCSI) sub-system, wherein the controller:

maintains the current queue depth values from the first processing state in response to not detecting a threshold change in relative demand for at least one VF;

in response to a threshold change in relative demand for at least one VF being detected, invokes, via an OS device driver, a request for a change in queue depth for each of at least one VF to correspond with the relative demands for request queue depth for respective virtual functions;

wherein the invoked request enables a completion of pending requests for a first processing state that uses one of (a) previous queue depth values corresponding to previous demands for request queue depth and (b) default queue depth values;

in response to successful completion of the first processing state, updating respective queue depth values within a Small Computer System Interface (SCSI) sub-system that provides a mechanism for transferring data between host devices and storage; and

in response to detecting the updates, initiates the second processing state in which request queues are resumed to continue processing requests.

3. The IHS of claim 1 , wherein the queue depth profiler represents a first component having a profiling capability and the OS device drivers represent additional components having corresponding profiling capability, wherein the profiling capability enables detection of queue full conditions on a per virtual function basis, wherein the queue depth profiler and the OS device drivers enable cooperative, collaborative processes that perform queue depth measurement functions and provide profiling measurements.

4. The IHS of claim 1 , wherein the controller:

in response to a first relative demand for a first virtual function exhibiting at least a threshold increase compared with a second relative demand for a second virtual function during queue depth profiling, provides a first desired queue depth value for the first virtual function and a second desired queue depth value for the second virtual function during the second processing state, wherein the first desired queue depth value represents a net increase of queue depth relative to the second desired queue depth.

5. The IHS of claim 1 , wherein the controller executes code of the queue depth profiler, which configures the controller to enable queue depth profiling to be initiated by one of: (a) dynamic, on-demand user activation; and (b) a pre-configured setting to enable automatic execution at periodic intervals.

6. The IHS of claim 1 , wherein the controller:

selects a time-based profiling window during which profiling measurements are performed, wherein a profiling window can be configured to accommodate various different types of workload processing operations.

7. The IHS of claim 1 , wherein the controller:

determines parameter values for average total outstanding requests, minimum total outstanding requests and maximum total outstanding requests for each virtual function during a profiling window period.

8. The IHS of claim 1 , wherein the controller:

reconfigures, in a converged infrastructure of the IHS, the queue depth of the storage controller to modify queue depths of corresponding host devices, while avoiding at least one of (i) re-initializing respective OS device drivers and (ii) rebooting respective host devices.

9. The IHS of claim 1 , wherein the controller:

applies a priority weighting to processing commands from each virtual function depending on at least one of: (a) a queue depth assigned to each virtual function; and (b) a number of pending commands from each virtual function.

10. In an information handling system (IHS), a processor-implemented method comprising:

determining, for each virtual function, a value for at least one parameter associated with unfulfilled, outstanding requests, the at least one parameter including a count of total outstanding requests;

comparing parameter values associated with outstanding requests across the plurality of virtual functions;

determining a relative demand for request queue depth for each respective virtual function based on the compared parameter values;

determining whether a threshold change in relative demand for at least one virtual function (VF) is detected; and

in response to a threshold change in relative demand for at least one VF being detected:

calculating for each virtual function, based on the relative demand, desired queue depth values each representing a maximum number of outstanding commands that can be allocated to a respective VF;

initiating a “quiesce I/O” mode and initiating completion of pending requests for a first processing state; and

in response to successful completion of the pending requests for the first processing state:

updating queue depth values from previous queue depth values to the desired queue depth values;

resuming I/O processing queues; and

initiating a second processing state to continue processing requests using the desired queue depth values.

11. The method of claim 10 , further comprising:

maintaining the current queue depth values from the first processing state in response to not detecting a threshold change in relative demand for at least one VF;

in response to a threshold change in relative demand for at least one VF being detected, invoking via an operating system (OS) device driver a request for a change in queue depth for each of at least one virtual function to correspond with the relative demands for request queue depth for respective virtual functions;

wherein the invoked request enables a completion of pending requests for a first processing state that uses one of (a) previous queue depth values corresponding to previous demands for request queue depth and (b) default queue depth values;

wherein in response to successful completion of the first processing state, providing updates to respective queue depth values within a Small Computer System Interface (SCSI) sub-system that provides a standard for transferring data between host devices and storage;

in response to providing the updates, initiating the second processing state in which request queues are resumed to continue processing requests.

12. The method of claim 10 , further comprising:

detecting queue full conditions on a per virtual function basis, wherein a queue depth profiler and OS device drivers enable cooperative, collaborative processes that perform queue depth measurement functions and provide profiling measurements.

13. The method of claim 10 , further comprising:

in response to a first relative demand for a first virtual function exhibiting at least a threshold increase compared with a second relative demand for a second virtual function during queue depth profiling, providing a first desired queue depth value for the first virtual function and a second desired queue depth value for the second virtual function during the second processing state, wherein the first desired queue depth value represents a net increase of queue depth relative to the second desired queue depth.

14. The method of claim 10 , further comprising:

enabling queue depth profiling, using a queue depth profiler, to be initiated by one of: (a) dynamic, on demand user activation; and (b) a pre-configured setting to enable automatic execution at periodic intervals.

15. The method of claim 10 , further comprising:

selecting a time-based profiling window during which profiling measurements are performed, wherein a profiling window can be configured to accommodate various different types of workload processing operations.

16. The method of claim 10 , further comprising:

determining parameter values for average total outstanding requests, minimum total outstanding requests and maximum total outstanding requests for each virtual function during a profiling window period.

17. The method of claim 10 , further comprising:

reconfiguring, in a converged, unified computing infrastructure of the IHS, the queue depth of the storage controller to modify queue depths of corresponding host devices while avoiding at least one of (i) re-initializing respective OS device drivers and (ii) rebooting respective host devices.

18. The method of claim 10 , further comprising:

applying a priority weighting to processing commands from each virtual function depending on at least one of: (a) a queue depth assigned to each virtual function; and (b) a number of pending commands from each virtual function.