System and method for reprioritizing high-latency input/output operations
A mechanism for reprioritizing high-latency input/output operations in a file system is provided. The mechanism expands a file access protocol, such as the direct access file system protocol, by including a hurry up command that adjusts the latency of a given input/output operation. The hurry up command can be employed in the Direct Access File System.
1. A method for modifying a latency of an input/output (I/O) request on a storage system, comprising:
receiving the I/O request from a computer, the request including a maximum latency value;
storing the I/O request in memory;
receiving a hurry up message from the computer, the hurry up message identifying the I/O request; and
changing, in response to the hurry up message, the maximum latency value of the I/O request to a reduced value.
2. The method of claim 1 , further comprising: storing the I/O request in a buffer in a volatile memory.
3. The method of claim 1 , further comprising: writing, in response to expiration of the reduced value, the I/O request to persistent storage.
4. The method of claim 3 , further comprising: sending, in response to writing the I/O request to persistent storage, an acknowledgment of the I/O request to the computer.
5. The method of claim 3 , further comprising: using a nonvolatile random access memory as the persistent storage.
6. The method of claim 3 further comprising: using disk storage as the persistent storage.
7. The method of claim 1 , further comprising: setting the reduced value to zero.
8. A computer, comprising:
a memory; and
a processor to,
i) receive an input/output (I/O) request, the request including a maximum latency value,
ii) store the I/O request in memory,
iii) receive a hurry up message identifying the I/O request, and
iv) change, in response to the hurry up message, the maximum latency value of the I/O request to a reduced value.
9. The computer of claim 8 , further comprising: a buffer in a volatile memory to store the I/O request.
10. The computer of claim 8 , further comprising:
a persistent storage,
the processor to write, in response to expiration of the reduced value, the I/O request to persistent storage.
11. The computer of claim 10 , further comprising: a processor to send, in response to writing the I/O request to persistent storage, an acknowledgment of the I/O request.
12. The computer of claim 10 , further comprising: a nonvolatile random access memory as the persistent storage.
13. The computer of claim 10 further comprising: a disk storage as the persistent storage.
14. The computer of claim 8 , further comprising: a value of zero as the reduced value.
15. A computer readable hardware media, comprising:
the computer readable hardware media containing instructions for the execution in a processor for the practice of the method of,
receiving an input/output (I/O) request from a computer, the request including a maximum latency value;
storing the I/O request in a memory;
receiving a hurry up message from the computer, the hurry up message identifying the I/O request; and
changing, in response to the hurry up message, the maximum latency value of the I/O request to a reduced value.
16. A method for modifying a latency of a message, comprising:
receiving a first message at a computer, the first message including an associated first latency value indicating a maximum latency period in which the first message should be acknowledged;
storing the first message and associated first latency value in a memory of the computer;
receiving a second message at the computer, the second message including a second latency value;
replacing the first latency value stored in the memory with the second latency value to thereby indicate a new maximum latency period in which the first message should be acknowledged.
17. The method of claim 16 , further comprising:
storing the first message in persistent memory; and
in response to storing the first message, sending an acknowledgment of the first message from the computer.
18. A computer, comprising:
a buffer for storing a first message including an associated first latency value indicating a maximum latency period in which the first message should be acknowledged; and
a processor configured to, in response to receiving a second message at the computer, the second message including a second latency value, replace the first latency value stored in the memory with the second latency value to thereby indicate a new maximum latency period in which the first message should be acknowledged.
19. The computer of claim 18 , further comprising:
a persistent memory;
the processor further configured to store the first message in persistent memory and in response to storing the first message, to send an acknowledgment of the first message.
20. A method for a server to modify a latency of an input/output (I/O) operation in response to a client, the method comprising:
receiving an I/O operation request at the server from the client, the request having a maximum latency value for the I/O operation;
receiving a hurry up message at the server from the client, the hurry up message requesting that the maximum latency value of the I/O operation be modified by the server; and
in response, modifying the maximum latency value of the I/O operation at the server.
21. The method as in claim 20 , wherein the hurry up message specifies an alternate latency value for the I/O operation.
22. The method as in claim 21 , further comprising:
modifying the maximum latency value of the I/O operation to the alternate latency value.
23. The method as in claim 21 , wherein the alternate latency value is a reduced value.
24. The method as in claim 21 , wherein the alternate latency value is zero.
25. The method as in claim 24 , further comprising:
in response to a zero latency value, substantially immediately performing the I/O operation.
26. The method as in claim 20 , further comprising:
performing the I/O operation no later than upon expiration of the maximum latency value of the I/O request.
27. The method as in claim 20 , further comprising:
sending an acknowledgment of the I/O operation to the client no later than upon expiration of the maximum latency value of the I/O request.
28. The method as in claim 20 , further comprising:
sending an acknowledgment of the I/O operation to the client upon completion of the I/O operation.
29. The method as in claim 20 , wherein the I/O request is a read request.
30. The method as in claim 20 , wherein the I/O request is a write request.
31. The method as in claim 30 , further comprising:
storing the received I/O operation in a volatile memory; and
writing the I/O operation to persistent storage no later than upon expiration of the maximum latency value.
32. The method as in claim 31 , wherein the persistent storage is a disk storage.
33. The method as in claim 31 , wherein the persistent storage is a nonvolatile random access memory (NVRAM) storage.
34. The method as in claim 33 , further comprising:
using the NVRAM persistent storage until the I/O operation is written to a slower persistent storage.
35. The method as in claim 20 , further comprising:
storing the received I/O operation in a buffer.
36. The method as in claim 35 , wherein the hurry up message specifies the I/O operation; the method further comprising:
locating the buffer associated with the I/O operation.
37. The method as in claim 36 , further comprising:
copying the located buffer to persistent storage no later than upon expiration of the maximum latency value; and
in response, sending an acknowledgment of the I/O operation to the client.
38. The method as in claim 20 , further comprising:
receiving the I/O request as a high-latency request.
39. The method as in claim 38 , further comprising:
modifying the maximum latency value of the I/O request to be a low-latency request.
40. A server computer, comprising:
a network adapter adapted to receive an operation request, the operation having an associated first latency value specifying a maximum latency period in which the operation should be acknowledged by the computer;
a buffer adapted to store the operation and associated first latency value;
the network adapter further adapted to receive a hurry up message having a second latency value and specifying the operation; and
a processor adapted to replace the first latency value stored in the buffer with the second latency value in response to a received hurry up message to thereby specify a New maximum latency period in which the operation should be acknowledged by the computer.
41. The server computer as in claim 40 , further comprising:
a persistent memory adapted to store the operation no later than upon expiration on the New maximum latency period; and
wherein the network adapter is further adapted to send an acknowledgment of the operation in response.
42. A method for controlling latency of execution of input/output requests in a server, the method comprising:
receiving an input/output operation request from a client, the input/output operation having an associated latency indicator;
receiving a latency control command from a client requesting a change of the latency indicator associated with the input/output operation; and
in response, changing the latency indicator associated with the input/output operation.
43. The method as in claim 42 , further comprising:
executing the input/output operation with a latency associated with the changed latency indicator.
44. The method as in claim 43 , wherein the latency indicator is associated with an execution priority relative to other pending input/output operations and the executing step executes the input/output operation responsive to the associated execution priority.
45. The method as in claim 42 , wherein the latency command comprises a hurry up command.
46. The method as in claim 42 , wherein the step of changing a latency indicator further comprises the step of changing the latency indicator to a selected value associated with a relative lower latency execution from a value associated with a relatively higher latency execution.
47. The method as in claim 46 , wherein the indicator is changed to a value under a preselected threshold associated with input/output operations requiring relatively low-latency execution.
48. The method as in claim 47 , wherein the threshold is dynamically set depending on queued input/output requests.
49. The method as in claim 47 , wherein the threshold is set dependent on the identity of the client.
50. The method as in claim 42 , wherein the latency indicator specifies a time within which execution of the input/output operation is requested.
51. The method as in claim 42 wherein the latency indicator is a field in the input/output operation request.
52. The method as in claim 51 , wherein the latency control command is received from the client subsequent to receipt of the input/output request.