Data replication using reinforcement learning to dynamically initiate transfer of replication data
Techniques are provided for data replication using reinforcement learning to dynamically initiate the transfer of replication data. One method comprises obtaining a state for transferring replication data from a first storage system to a second storage system, wherein the state comprises an amount of replication data to be transferred to the second storage system; assigning a reward value to previously completed replication sessions based on whether a respective previously completed replication session satisfies a designated data recovery objective; determining a start time, using a reinforcement learning framework, for transferring the replication data to the second storage system, wherein the determining is based on the assigned reward values; and initiating the transfer of the replication data to the second storage system using the determined start time. The reward value for a given concurrent replication session may be based on respective replication data transfer times of multiple concurrent replication sessions.
1. A method, comprising:
obtaining a state of an information technology infrastructure for transferring replication data from a first storage system to a second storage system, wherein the state comprises an amount of replication data to be transferred to the second storage system;
assigning a reward value to one or more previously completed replication sessions based at least in part on whether a respective previously completed replication session satisfies a designated data recovery objective;
determining a start time, using a reinforcement learning framework, for transferring the replication data to the second storage system, wherein the determining is based at least in part on one or more of the assigned reward values; and
initiating the transfer of the replication data from the first storage system to the second storage system using the determined start time;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
2. The method of claim 1 , wherein the reinforcement learning framework determines the start time using an action-value function, wherein the action-value function (i) characterizes an expected return for a given action and (ii) is based at least in part on the one or more reward values.
3. The method of claim 1 , wherein the amount of replication data to be transferred to the second storage system comprises a differential amount of data relative to a prior replication data transfer.
4. The method of claim 3 , wherein the differential amount of data relative to the prior replication data transfer is determined by comparing a refreshed snapshot to a base snapshot.
5. The method of claim 1 , wherein the state further comprises a plurality of concurrent replication sessions, and wherein the reward value for a given concurrent replication session is based at least in part on respective replication data transfer times of at least some of the plurality of concurrent replication sessions.
6. The method of claim 1 , wherein the assigning the reward value to the one or more previously completed replication sessions is based at least in part on an evaluation of a replication data accumulation time and a replication data transfer time.
7. The method of claim 1 , wherein the designated data recovery objective specifies a limit on a time difference between a beginning of a first replication transfer and an end of a second replication transfer.
8. The method of claim 1 , wherein the designated data recovery objective is based at least in part on a designated amount of lost data, measured in units of time, due to a failure of one or more information technology assets in the information technology infrastructure.
9. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory, wherein the memory comprises program code;
the at least one processing device being configured to execute the program code to cause the at least one processing device to implement the following steps:
obtaining a state of an information technology infrastructure for transferring replication data from a first storage system to a second storage system, wherein the state comprises an amount of replication data to be transferred to the second storage system;
assigning a reward value to one or more previously completed replication sessions based at least in part on whether a respective previously completed replication session satisfies a designated data recovery objective;
determining a start time, using a reinforcement learning framework, for transferring the replication data to the second storage system, wherein the determining is based at least in part on one or more of the assigned reward values; and
initiating the transfer of the replication data from the first storage system to the second storage system using the determined start time.
10. The apparatus of claim 9 , wherein the reinforcement learning framework determines the start time using an action-value function, wherein the action-value function (i) characterizes an expected return for a given action and (ii) is based at least in part on the one or more reward values.
11. The apparatus of claim 9 , wherein the amount of replication data to be transferred to the second storage system comprises a differential amount of data relative to a prior replication data transfer, wherein the differential amount of data relative to the prior replication data transfer is determined by comparing a refreshed snapshot to a base snapshot.
12. The apparatus of claim 9 , wherein the state further comprises a plurality of concurrent replication sessions, and wherein the reward value for a given concurrent replication session is based at least in part on respective replication data transfer times of at least some of the plurality of concurrent replication sessions.
13. The apparatus of claim 9 , wherein the assigning the reward value to the one or more previously completed replication sessions is based at least in part on an evaluation of a replication data accumulation time and a replication data transfer time.
14. The apparatus of claim 9 , wherein the designated data recovery objective is based at least in part on a designated amount of lost data, measured in units of time, due to a failure of one or more information technology assets in the information technology infrastructure.
15. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:
obtaining a state of an information technology infrastructure for transferring replication data from a first storage system to a second storage system, wherein the state comprises an amount of replication data to be transferred to the second storage system;
assigning a reward value to one or more previously completed replication sessions based at least in part on whether a respective previously completed replication session satisfies a designated data recovery objective;
determining a start time, using a reinforcement learning framework, for transferring the replication data to the second storage system, wherein the determining is based at least in part on one or more of the assigned reward values; and
initiating the transfer of the replication data from the first storage system to the second storage system using the determined start time.
16. The non-transitory processor-readable storage medium of claim 15 , wherein the designated data recovery objective is based at least in part on a designated amount of lost data, measured in units of time, due to a failure of one or more information technology assets in the information technology infrastructure.
17. The non-transitory processor-readable storage medium of claim 15 , wherein the assigning the reward value to the one or more previously completed replication sessions is based at least in part on an evaluation of a replication data accumulation time and a replication data transfer time.
18. The non-transitory processor-readable storage medium of claim 15 , wherein the reinforcement learning framework determines the start time using an action-value function, wherein the action-value function (i) characterizes an expected return for a given action and (ii) is based at least in part on the one or more reward values.
19. The non-transitory processor-readable storage medium of claim 15 , wherein the amount of replication data to be transferred to the second storage system comprises a differential amount of data relative to a prior replication data transfer, wherein the differential amount of data relative to the prior replication data transfer is determined by comparing a refreshed snapshot to a base snapshot.
20. The non-transitory processor-readable storage medium of claim 15 , wherein the state further comprises a plurality of concurrent replication sessions, and wherein the reward value for a given concurrent replication session is based at least in part on respective replication data transfer times of at least some of the plurality of concurrent replication sessions.