Filtered data cache eviction
A filtered data cache eviction method preserves data that might otherwise be evicted by condensing the data into fewer containers. In particular, hot pages or pages that are associated with a particular application's working set of data are condensed into fewer containers rather than being evicted. The data that is copy forwarded includes blocks or pages of data tracked as having been recently and/or frequently accessed or otherwise associated with an active file.
1. A computer-implemented method for preserving data cached in a cache memory device undergoing eviction, the method comprising:
storing one or more blocks of data in a data cache in a cache memory device, the data cache composed of data structures in which some of the one or more blocks of data is stored;
tracking how frequently and how recently data stored in the data cache is accessed over time;
determining whether contents of the data structures to be evicted can be condensed to preserve data tracked as having been any one of frequently and recently accessed; and
condensing the contents of the data structures to be evicted to reduce a larger number of data structures containing the data into a fewer number of data structures by copy-forwarding the contents from less populated data structures to other data structures that are equivalently or more populated, the condensed contents of the data structures including the data tracked as having been any one of frequently and recently accessed.
2. The method of claim 1 , wherein:
the data structures are write evict units (WEUs) in the cache memory device that each have a size determined based on an erase unit size of the cache memory device; and
data stored in the cache memory device is erased on a per-erase unit basis according to a specification of the cache memory device.
3. The method of claim 1 , wherein determining whether contents of the data structures can be condensed to filter data tracked as having been any one of frequently and recently accessed further comprises:
computing for each data structure averages of how frequently and how recently any one or more blocks of data stored in the data structure were accessed over time;
sorting the data structures based on the computed averages; and
copy forwarding from less populated data structures to more populated data structures those blocks of data tracked as having been any of more frequently and more recently accessed as compared to other blocks of data.
4. The method of claim 1 , further comprising:
tracking whether data stored in the data cache belongs to an application's working set of data; and
determining whether data stored in the data cache can be condensed to preserve the application's working set of data.
5. The method of claim 4 , wherein tracking whether data stored in the data cache belongs to an application's working set of data includes determining whether the data is associated with an active or inactive file of an application, the data associated with the active file indicating that the data belongs to the application's working set of data.
6. The method of claim 1 , wherein the one or more blocks of data are any one of a page and a segment capable of being identified in a map associating each block of data to the data structure in which it is stored, the method further comprising:
establishing counters for each block of data to contain cache hit counts, the cache hit counts indicating how frequently and how recently each block was accessed over time.
7. A storage system for caching data using a cache memory device, comprising:
a cache memory device having a data cache in which to store one or more blocks of data over time, the data cache composed of data structures each capable of containing some of the one or more blocks of data; and
a cache manager executed by a processor to:
store data in the data cache in the cache memory device, the data cache composed of data structures in which data is stored;
track how frequently and how recently data stored in the data cache is accessed over time;
determine whether contents of the data structures to be evicted can be condensed to preserve data tracked as having been any one of frequently and recently accessed; and
condense the contents of the data structures to be evicted to reduce a larger number of data structures containing the data into a fewer number of data structures by copy-forwarding the contents from less populated data structures to other data structures that are equivalently or more populated, the condensed contents of the data structures to include the data tracked as having been any one of frequently and recently accessed.
8. The storage system of claim 7 , wherein:
the data structures are write evict units (WEUs) in the cache memory device that each have a size determined based on an erase unit size of the cache memory device; and
data stored in the cache memory device is erased on a per-erase unit basis according to a specification of the cache memory device.
9. The storage system of claim 7 , wherein to determine whether contents of the data structures can be condensed to preserve data tracked as having been any one of frequently and recently accessed, the cache manager executed by the processor is further to:
compute for each data structure averages of how frequently and how recently any one or more blocks of data stored in the data structure were accessed over time;
sort the data structures based on the computed averages; and
copy forward from less populated data structures to more populated data structures those blocks of data tracked as having been any of more frequently and more recently accessed as compared to other blocks of data.
10. The storage system of claim 7 , wherein the cache manager executed by the processor is further to:
track whether data stored in the data cache belongs to an application's working set of data; and
determine whether data stored in the data cache can be condensed to preserve the application's working set of data.
11. The storage system of claim 10 , wherein to track whether data stored in the data cache belongs to an application's working set of data, the cache manager executed by the processor is further to:
determine whether the data is associated with an active or inactive file of an application, the data associated with the active file indicating that the data belongs to the application's working set of data.
12. The storage system of claim 7 , wherein the one or more blocks of data are any one of a page and a segment capable of being identified in a map associating each block of data to the data structure in which it is stored, and the cache manager executed by the processor is further to:
establish counters for each block of data to contain cache hit counts, the cache hit counts indicating how frequently and how recently each block was accessed over time.
13. A non-transitory computer-readable storage medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations for managing a data cache in a cache memory device, the operations comprising:
storing one or more blocks of data in a data cache in a cache memory device, the data cache composed of data structures in which some of the one or more blocks of data is stored;
tracking how frequently and how recently data stored in the data cache is accessed over time;
determining whether contents of the data structures to be evicted can be condensed to preserve data tracked as having been any one of frequently and recently accessed; and
condensing the contents of the data structures to be evicted to reduce a larger number of data structures containing the data into a fewer number of data structures by copy-forwarding the contents from less populated data structures to other data structures that are equivalently or more populated, the condensed contents of the data structures including the data tracked as having been any one of frequently and recently accessed.
14. The non-transitory computer-readable storage medium of claim 13 , wherein:
the data structures are write evict units (WEUs) in the cache memory device that each have a size determined based on an erase unit size of the cache memory device; and
data stored in the cache memory device is erased on a per-erase unit basis according to a specification of the cache memory device.
15. The non-transitory computer-readable storage medium of claim 13 , wherein determining whether contents of the data structures can be condensed to preserve data tracked as having been any one of frequently and recently accessed further includes operations for:
computing for each data structure averages of how frequently and how recently any one or more blocks of data stored in the data structure were accessed over time;
sorting the data structures based on the computed averages; and
copy forwarding from less populated data structures to more populated data structures those blocks of data tracked has having been any of more frequently and more recently accessed as compared to other blocks of data.
16. The non-transitory computer-readable storage medium of claim 13 , wherein determining whether contents of the data structures can be condensed to preserve data tracked as having been any one of frequently and recently accessed further includes operations for:
tracking whether data stored in the data cache belongs to an application's working set of data; and
determining whether data stored in the data cache can be condensed to preserve the application's working set of data.
17. The non-transitory computer-readable storage medium of claim 16 , wherein the operation for tracking whether data stored in the data cache belongs to an application's working set of data further includes an operation for determining whether the data is associated with an active or inactive file of an application, the data associated with the active file indicating that the data belongs to the application's working set of data.
18. The non-transitory computer-readable storage medium of claim 13 , wherein the one or more blocks of data are any one of a page and a segment capable of being identified in a map associating each block of data to the data structure in which it is stored, and the operations further include an operation for establishing counters for each block of data to contain cache hit counts, the cache hit counts indicating how frequently and how recently each block was accessed over time.