Optimizing I/O latency by software stack latency reduction in a cooperative thread processing model
A method for use in a storage node, the method comprising: instantiating a first poller for detecting whether pending storage device operations have been completed; executing the first poller to identify a first storage device operation that has been completed, wherein executing the first poller includes: (a) executing a first function to detect whether a completion queue corresponding to a storage device driver is empty, the first function being arranged to read a content of a memory location that is associated with the completion queue, (b) terminating the execution of the first poller when the completion queue is empty, and (c) executing a system call function to the storage device driver when the completion queue is non-empty; identifying a first thread that is waiting for the first storage device operation to be completed; and transitioning the first thread from a waiting state to a ready state.
1. A method for use in a storage node, the method comprising:
instantiating a first poller for detecting whether pending storage device operations have been completed;
executing the first poller to identify a first storage device operation that has been completed, wherein executing the first poller includes: (a) executing a first function to detect whether a completion queue corresponding to a storage device driver is empty, the first function being arranged to read a content of a memory location that is associated with the completion queue, the first function being executed in a user space of the storage node, (b) terminating the execution of the first poller when the completion queue is empty, and (c) executing a system call function to the storage device driver only when the completion queue is not empty, wherein the system call function is executed in a kernel space of the storage node, and the system call function is configured to identify one or more operations that are listed in the completion queue;
identifying a first thread that is waiting for the first storage device operation to be completed; and
transitioning the first thread from a waiting state to a ready state.
2. The method of claim 1 , wherein the first thread is arranged to execute a processor yield operation at intervals that are less than or equal to a scheduling interval of the first poller.
3. The method of claim 1 , further comprising:
instantiating a second poller for detecting whether pending storage device operations have been completed;
executing the second poller to identify a second storage device operation that has been completed, wherein executing the second poller includes: (a) executing the first function to detect whether the completion queue is empty, (b) terminating the execution of the second poller when the completion queue is empty, and (c) executing the system call function to the storage device driver when the completion queue is not empty;
identifying a second thread that is waiting for the second storage device operations to be completed; and
transitioning the second thread from a waiting state to a ready state,
wherein the first poller is executed by a first processor of the storage node and the second poller is executed by a second processor of the storage node.
4. The method of claim 3 , wherein the first poller and the second poller are executed concurrently, the first thread is executed on the first processor, and the second thread is executed on the second processor.
5. The method of claim 3 , further comprising instantiating a poller synchronization object, wherein the first poller and the second poller are executed at least in part sequentially based on the poller synchronization object, the first thread is executed on either one of the first processor and the second processor, and the second thread is executed on either one of the first processor and the second processor.
6. An apparatus, comprising:
a memory; and
a processing circuitry operatively coupled to the memory, the processing circuitry being configured to perform the operations of:
instantiating a first poller for detecting whether pending storage device operations have been completed;
executing the first poller to identify a first storage device operation that has been completed, wherein executing the first poller includes: (a) executing a first function to detect whether a completion queue corresponding to a storage device driver is empty, the first function being arranged to read a content of a memory location that is associated with the completion queue, the first function being executed in a user space of the storage node, (b) terminating the execution of the first poller when the completion queue is empty, and (c) executing a system call function to the storage device driver only when the completion queue is not empty, wherein the system call function is executed in a kernel space of the storage node, and the system call function is configured to identify one or more operations that are listed in the completion queue;
identifying a first thread that is waiting for the first storage device operation to be completed; and
transitioning the first thread from a waiting state to a ready state.
7. The apparatus of claim 6 , wherein the first thread is arranged to execute a processor yield operation at intervals that are less than or equal to a scheduling interval of the first poller.
8. The apparatus of claim 6 , wherein the processing circuitry is further configured to perform the operations of:
instantiating a second poller for detecting whether pending storage device operations have been completed;
executing the second poller to identify a second storage device operation that has been completed, wherein executing the second poller includes: (a) executing the first function to detect whether the completion queue is empty, (b) terminating the execution of the second poller when the completion queue is empty, and (c) executing the system call function to the storage device driver when the completion queue is not empty;
identifying a second thread that is waiting for the second storage device operation to be completed; and
transitioning the second thread from a waiting state to a ready state,
wherein the processing circuitry includes a first processor and a second processor, and
wherein the first poller is executed by the first processor and the second poller is executed by the second processor.
9. The apparatus of claim 8 , wherein the first poller and the second poller are executed concurrently, the first threads is executed on the first processor, and the second thread is executed on the second processor.
10. The apparatus of claim 8 , further comprising instantiating a poller synchronization object, wherein the first poller and the second poller are executed at least in part sequentially based on the poller synchronization object, the first thread is executed on either one of the first processor and the second processor, and the second thread is executed on either one of the first processor and the second processor.
11. A non-transitory computer-readable medium storing one or more processor-executable instructions, which when executed by a processing circuitry of a storage node cause the processing circuitry to perform the operations of:
instantiating a first poller for detecting whether pending storage device operations have been completed;
executing the first poller to identify a first storage device operation that has been completed, wherein executing the first poller includes: (a) executing a first function to detect whether a completion queue corresponding to a storage device driver is empty, the first function being arranged to read a content of a memory location that is associated with the completion queue, the first function being executed in a user space of the storage node, (b) terminating the execution of the first poller when the completion queue is empty, and (c) executing a system call function to the storage device driver when the completion queue is not empty, wherein the first poller is configured to execute the system call function only when the completion queue is not empty, the system call function is executed in a kernel space of the storage node, and the system call function is configured to identify one or more operations that are listed in the completion queue;
identifying a first thread that is waiting for the first storage device operation to be completed; and
transitioning the first thread from a waiting state to a ready state.
12. The non-transitory computer-readable of claim 11 , wherein the first thread is arranged to execute a processor yield operation at intervals that are less than or equal to a scheduling interval of the first poller.
13. The non-transitory computer-readable of claim 11 , further comprising:
instantiating a second poller for detecting whether pending storage device operations have been completed;
executing the second poller to identify one or more second storage device operations that have been completed, wherein executing the second poller includes: (a) executing the first function to detect whether the completion queue is empty, (b) terminating the execution of the second poller when the completion queue is empty, and (c) executing the system call function to the storage device driver when the completion queue is not empty;
identifying a second thread that is waiting for the second storage device operation to be completed; and
transitioning the second thread from a waiting state to a ready state,
wherein the processing circuitry includes a first processor and a second processor, and
wherein the first poller is executed by the first processor and the second poller is executed by the second processor.
14. The non-transitory computer-readable of claim 13 , wherein the first poller and the second poller are executed concurrently, the first thread is executed on the first processor, and the second thread is executed on the second processor.
15. The non-transitory computer-readable of claim 13 , further comprising instantiating a poller synchronization object, wherein the first poller and the second poller are executed at least in part sequentially based on the poller synchronization object, the first thread include is executed on either one of the first processor and the second processor, and the second thread is executed on either one of the first processor and the second processor.