Shaping I/O traffic by managing queue depth in fractional increments
View Patent ↗A method for managing input/output (I/O) traffic in an information handling system. The method may include receiving electronic I/O requests from a network-attached server, determining a queue depth limit, monitoring latency of processed electronic I/O requests, and processing received electronic I/O requests. The number of electronic I/O requests permitted to be processed over a period of time may be based on a mathematical combination of the queue depth limit and a latency of processed electronic I/O requests. The determined queue depth limit may be a fractional value.
1. A method for managing input/output (I/O) traffic in an information handling system, the method comprising:
monitoring latency of previously processed electronic I/O requests;
determining a queue depth limit;
receiving additional electronic I/O requests from a network-attached server; and
processing the additional electronic I/O requests, wherein the number of additional electronic I/O requests permitted to be processed over a period of time is based on a mathematical combination of the queue depth limit and the latency of previously processed electronic I/O requests.
2. The method of claim 1 , wherein processing the additional electronic I/O requests comprises:
managing a virtual bucket comprising a number of virtual tokens, each token representative of authorization for an I/O request to be processed; and
permitting one of the additional electronic I/O requests to be processed if the virtual bucket comprises a number of virtual tokens greater than zero, and decrementing the number of virtual tokens by at least one.
3. The method of claim 2 , wherein calculating a number of virtual tokens to add to the virtual bucket at any given time t x is performed according to the equation:
Tokens( t x )=QueueDepthLimit*TimeElapsed( t x −t x-1 )/Latency
wherein Tokens(t x ) gives a value corresponding to a number of virtual tokens to be added to the virtual bucket at time t x ; QueueDepthLimit is the determined queue depth limit; TimeElapsed(t x −t x-1 ) is an amount of time elapsed between time t x and a time t x-1 , where t x-1 is a point in time prior to time t x ; and Latency is a latency of previously processed electronic I/O requests that have completed during the time elapsed between time t x and a time t x-1 .
4. The method of claim 3 , wherein the QueueDepthLimit is a fractional value.
5. The method of claim 2 , wherein processing the additional electronic I/O requests further comprising waiting to process a received I/O request if the virtual bucket comprises a number of virtual tokens no more than zero until the virtual bucket receives additional virtual tokens.
6. The method of claim 1 , wherein the mathematical combination of the queue depth limit and the latency of previously processed electronic I/O requests is calculated at periodic intervals.
7. The method of claim 6 , wherein the mathematical combination of the queue depth limit and the latency of previously processed electronic I/O requests is additionally calculated when an electronic I/O request is received.
8. The method of claim 1 , wherein the mathematical combination of the queue depth limit and the latency of previously processed electronic I/O requests is calculated at random intervals.
9. The method of claim 1 , wherein the mathematical combination of the queue depth limit and the latency of previously processed electronic I/O requests is calculated when an electronic I/O request is received.
10. The method of claim 1 , wherein the queue depth limit is a fractional value.
11. A method for managing input/output (I/O) traffic in an information handling system, the method comprising:
monitoring latency of previously processed electronic I/O requests, the electronic I/O requests having been received from a plurality of network-attached servers;
determining a queue depth limit for each of the plurality of network-attached servers;
receiving additional electronic I/O requests from the plurality of network-attached servers; and
processing the additional electronic I/O requests for each of the network-attached servers, wherein the number of additional electronic I/O requests permitted to be processed for a given network-attached server over a period of time is based on a mathematical combination of the queue depth limit determined for that network-attached server and the latency of previously processed electronic I/O requests.
12. The method of claim 11 , wherein the queue depth limit for one of the plurality of network-attached servers is different than at least one other of the plurality of network-attached servers.
13. The method of claim 11 , wherein processing received I/O requests for each of the network-attached servers comprises, for each of the network-attached servers:
managing a virtual bucket comprising a number of virtual tokens, each token representative of authorization for an I/O request from at least that network-attached server to be processed; and
permitting one of the additional electronic I/O request from that network-attached server to be processed if the virtual bucket comprises a number of virtual tokens greater than zero, and decrementing the number of virtual tokens by at least one.
14. The method of claim 11 , wherein the mathematical combination is calculated for a given network-attached server when an I/O request is received from that network-attached server.
15. The method of claim 11 , wherein the mathematical combination is calculated for each network-attached server at periodic intervals.
16. The method of claim 15 , wherein calculating a number of virtual tokens to add to the virtual bucket at any given time t x is performed according to the equation:
Tokens( t x )=QueueDepthLimit*TimeElapsed( t x −t x-1 )/Latency
wherein Tokens(t x ) gives a value corresponding to a number of virtual tokens to be added to the virtual bucket at time t x ; QueueDepthLimit is the determined queue depth limit for that network-attached server; TimeElapsed(t x −t x-1 ) is an amount of time elapsed between time t x and a time t x-1 , where t x-1 is a point in time prior to time t x ; and Latency is a latency of previously processed electronic I/O requests that have completed during the time elapsed between time t x and a time t x-1 .
17. The method of claim 16 , wherein the QueueDepthLimit for at least one of the network-attached servers is a fractional value.
18. The method of claim 11 , wherein the queue depth limit for at least one of the network-attached servers is a fractional value.
19. An information handling system comprising:
a data storage system receiving I/O requests from two or more network-attached servers; and
a controller managing the data storage system, including:
monitoring latency of previously processed electronic I/O requests;
storing a queue depth limit for each of the two or more network-attached servers;
receiving additional electronic I/O requests from the two or more network-attached servers; and
processing the additional electronic I/O requests for each of the network-attached servers, wherein the number of additional electronic I/O requests permitted to be processed for a given network-attached server over a period of time is based on a mathematical combination of the queue depth limit determined for that network-attached server and the latency of previously processed electronic I/O requests.
20. The information handling system of claim 19 , wherein the queue depth limit for at least one of the two or more network-attached servers is a fractional value.