Prioritization of virtual volumes to take offline in a thin provisioning system

Abstract

Prioritizing virtual volumes to take offline in a thin provisioning system with garbage collection. The method categorizes virtual volumes based on garbage collection properties of their write behavior and adds metadata indicating a category of a virtual volume. The method schedules virtual volumes to be taken offline by predicting virtual volume space utilization of active virtual volumes for a defined time period in combination with estimated garbage collection in that period to determine a need to take virtual volumes offline. The method selects virtual volumes to take offline by their category to ensure that the virtual volumes producing the most garbage collection unfriendly workloads are taken offline first.

1 . A method comprising:
 receiving a virtual volume data set including information indicative of an identity of a plurality of virtual volumes and write behavior respectively corresponding to the virtual volumes of the plurality of virtual volumes;   for each given virtual volume of the plurality of virtual volumes, determining, by machine logic based rules, a garbage collection category, of a plurality of predetermined garbage collection categories, corresponding to the given virtual volume, with the determination being based, at least in part, upon write behavior of the given virtual volume, with the predetermined garbage collection categories corresponding to relative levels of garbage collection friendliness with the relative levels of garbage collection friendliness being values that respectively correspond to a plurality of workloads, with workloads characterized by a relatively large amount of expected rewrite I/O (input/output) having relatively unfavorable garbage collection friendliness values, and with workloads having relatively small amounts of expected rewrite I/O having relatively favorable garbage collection friendliness values;   for each given virtual volume of the plurality of virtual volumes, adding garbage collection metadata to the given virtual volume, with the metadata including information indicative of the garbage collection category of the virtual volume;   adding one or more levels of priority metadata to virtual volumes to exclude or defer priority virtual volumes from being selected as selected virtual volume(s) to take offline;   sorting virtual volumes within a selected garbage collection category of the plurality of garbage collection categories to select high space utilization volumes to be taken offline first with the high space utilization volumes being virtual volumes that includes an amount of data that occupies a relatively large amount of data storage space inside a virtual disk;   selecting, by machine logic based rules, selected virtual volume(s), with the selection being based, at least in part, upon garbage collection categories indicated by the garbage collection metadata of the plurality of virtual volumes;   selectively taking the selected virtual volume(s) offline, while maintaining virtual volume(s) of the plurality of virtual volumes other than the selected virtual volume(s) online;   estimating, by machine logic rules, a data volume of garbage collection for a predetermined time period to determine estimated garbage collection values during the predetermined time period;   predicting, by machine logic rules, predicted virtual volume space utilization of active virtual volume(s) of the plurality of virtual volumes over the predetermined time period; and   scheduling, by machine logic rules, the taking offline of the selected virtual volume(s) over the predetermined time period based, at least in part, upon the estimated garbage collection values and the predicted virtual volume space utilization;   wherein:
 the predetermined garbage collection categories are based on a combination of a level of random or sequential write workloads and a level of new writes or rewrites, with a high level of random write workload with a low level of rewrite being a level of write workload and rewrite that causes the forward lookup metadata to be updated in an inefficient manner, and with the high level of random write workload with the low level of rewrite being a garbage collection unfriendly workload; 
 the selection of virtual volumes to take offline includes categories of virtual volumes in the order: 
 virtual volumes producing a high level of new writes, virtual volumes producing garbage collection unfriendly write workloads, virtual volumes producing garbage collection friendly write workloads, with the garbage collection friendly write workloads being workloads with a level of sequential write workloads that requires the forward lookup metadata to be updated in an efficient manner, and virtual volumes marked as priority volumes; or 
 virtual volumes with a high level of random writes and a low level of rewrites, virtual volumes with a high level of sequential writes and a low level of rewrites, virtual volumes with a high level of random writes and a high level of rewrites, and virtual volumes with a high level of sequential writes and a high level of rewrites.