Intelligent cache pre-fetch
An intelligent cache pre-fetch system includes a pre-fetch throttling scheme to monitor a cache hit rate context. Pre-fetch reads of additional data are only launched when the context is below a given threshold. A pre-fetch read of additional data can be selectively initiated after determining that references to neighboring segments related to a compression region already in memory are not yet present in the cache. Additional throttling of pre-fetch reads can be accomplished by only initiating the selective pre-fetch of additional data after determining whether the compression region to which the neighboring segments are related is a hot region, where a hot region is characterized as a compression region having data that is accessed frequently as compared to data in other compression regions.
1. A computer-implemented method for managing read operations in a tiered data storage system, the method comprising:
storing compressions regions in a data cache in a cache memory device, the compression regions containing data accessed by applications;
determining that additional data related to data contained in any one or more of the compression regions is not yet present in the data cache;
initiating a pre-fetch read of the additional data from a storage unit;
completing the pre-fetch read for a hot region, the hot region characterized by compression regions containing any one or more of frequently and recently accessed data; and
stopping the pre-fetch read for a cold region to avoid unnecessary disk Input/Output operations that would be incurred by completing the pre-fetch read, wherein the cold region is characterized by compression regions containing neither frequently nor recently accessed data.
2. The computer-implemented method of claim 1 , wherein determining that additional data related to data contained in any one or more of the compression regions is not yet present in the data cache is based on metadata associated with the compression regions, including:
receiving a reference for a compression region in a callback of a read operation for the compression region, the callback containing the reference as one or more contiguous references associated with data contained in the compression region;
selecting metadata for the reference, the metadata including a fingerprint of the reference; and
performing a cache-only lookup of the reference using the fingerprint to determine whether the additional data is not yet present in the data cache.
3. The computer-implemented method of claim 1 , further comprising:
monitoring a cache hit rate context of the pre-fetch read;
determining that the cache hit rate context has met a threshold after which pre-fetch reads are throttled;
stopping the pre-fetch read to avoid unnecessary disk Input/Output operations that would be incurred by completing the pre-fetch read.
4. The computer-implemented method of claim 3 , wherein monitoring a cache hit rate context of the pre-fetch read includes:
initializing a cache miss count and a cache hit count over a window of time;
accumulating the cache miss count and the cache hit count over the window of time; and
computing the cache hit rate context for the window of time based on a percentage of cache hit count relative to the cache miss count and the cache hit count combined.
5. The computer-implemented method of claim 3 , wherein the threshold after which pre-fetch reads are throttled varies depending on any one or more of a disk I/O capacity of the tiered data storage system and a current cache hit rate context.
6. The computer-implemented method of claim 3 , wherein the threshold after which pre-fetch reads are throttled varies depending on a temporal locality associated with the compression regions stored in the data cache, wherein a high temporal locality increases the threshold and a low temporal locality lowers the threshold.
7. The computer-implemented method of claim 1 , wherein the cache memory device is any one of a solid-state device and a flash memory device.
8. A storage system for managing read operations in a tiered data storage system, comprising:
a cache memory device having a data cache in which to store one or more compressions regions, the compression regions containing data accessed by applications;
a cache manager executed by a processor to:
determine that additional data related to data contained in any one or more of the compression regions is not yet present in the data cache;
initiate a pre-fetch read of the additional data from a storage unit;
complete the pre-fetch read for a hot region, the hot region characterized by compression regions containing any one or more of frequently and recently accessed data; and
stop the pre-fetch read for a cold region to avoid unnecessary disk Input/Output operations that would be incurred by completing the pre-fetch read, wherein the cold region is characterized by compression regions containing neither frequently nor recently accessed data.
9. The storage system of claim 8 , wherein to determine that additional data related to data contained in any one or more of the compression regions is not yet present in the data cache is based on metadata associated with the compression regions, the cache manager executed by the processor is further to:
receive a reference for a compression region in a callback of a read operation for the compression region, the callback containing the reference as one or more contiguous references associated with data contained in the compression region;
select metadata for the reference, the metadata including a fingerprint of the reference; and
perform a cache-only lookup of the reference using the fingerprint to determine whether the additional data is not yet present in the data cache.
10. The storage system of claim 8 , the cache manager executed by the processor is further to:
monitor a cache hit rate context of the pre-fetch read;
determine that the cache hit rate context has met a threshold after which pre-fetch reads are throttled; and
stop the pre-fetch read to avoid unnecessary disk Input/Output operations that would be incurred by completing the pre-fetch read.
11. The storage system of claim 10 , wherein to monitor a cache hit rate context of the pre-fetch read, the cache manager executed by the processor is further to:
initialize a cache miss count and a cache hit count over a window of time;
accumulate the cache miss count and the cache hit count over the window of time; and
compute the cache hit rate context for the window of time based on a percentage of cache hit count relative to the cache miss count and the cache hit count combined.
12. The storage system of claim 10 , wherein the threshold after which pre-fetch reads are throttled varies depending on any one or more of a disk I/O capacity of the tiered data storage system and a current cache hit rate context.
13. The storage system of claim 10 , wherein the threshold after which pre-fetch reads are throttled varies depending on a temporal locality associated with the compression regions stored in the data cache, wherein a high temporal locality increases the threshold and a low temporal locality lowers the threshold.
14. The storage system of claim 10 , wherein the cache memory device is any one of a solid-state device and a flash memory device.
15. 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 read operations in a tiered data storage system, the operations comprising:
storing compressions regions in a data cache in a cache memory device, the compression regions containing data accessed by applications;
determining that additional data related to data contained in any one or more of the compression regions is not yet present in the data cache;
initiating a pre-fetch read of the additional data from a storage unit;
completing the pre-fetch read for a hot region, the hot region characterized by compression regions containing any one or more of frequently and recently accessed data; and
stopping the pre-fetch read for a cold region to avoid unnecessary disk Input/Output operations that would be incurred by completing the pre-fetch read, wherein the cold region is characterized by compression regions containing neither frequently nor recently accessed data.
16. The non-transitory computer-readable storage medium of claim 15 , wherein the operation for determining that additional data related to data contained in any one or more of the compression regions is not yet present in the data cache is based on metadata associated with the compression regions, further includes operations for:
receiving a reference for a compression region in a callback of a read operation for the compression region, the callback containing the reference as one or more contiguous references associated with data contained in the compression region;
selecting metadata for the reference, the metadata including a fingerprint of the reference; and
performing a cache-only lookup of the reference using the fingerprint to determine whether the additional data is not yet present in the data cache.
17. The non-transitory computer-readable storage medium of claim 15 , the operations comprising:
monitoring a cache hit rate context of the pre-fetch read;
determining that the cache hit rate context has met a threshold after which pre-fetch reads are throttled;
stopping the pre-fetch read to avoid unnecessary disk Input/Output operations that would be incurred by completing the pre-fetch read.
18. The non-transitory computer-readable storage medium of claim 15 , wherein the operation to monitor a cache hit rate context of the pre-fetch read further includes operations for:
initializing a cache miss count and a cache hit count over a window of time;
accumulating the cache miss count and the cache hit count over the window of time; and
computing the cache hit rate context for the window of time based on a percentage of cache hit count relative to the cache miss count and the cache hit count combined.
19. The non-transitory computer-readable storage medium of claim 18 , wherein the threshold after which pre-fetch reads are throttled varies depending on any one or more of a disk I/O capacity of the tiered data storage system and a current cache hit rate context.
20. The non-transitory computer-readable storage medium of claim 18 , wherein the threshold after which pre-fetch reads are throttled varies depending on a temporal locality associated with the compression regions stored in the data cache, wherein a high temporal locality increases the threshold and a low temporal locality lowers the threshold.