Methods for improved data caching
View Patent ↗Write transactions with large amounts of data using a typical cache may consume over half of the available backing store bandwidth because of the way traditional caching algorithms fill lines during a write-invoked eviction. Relaxing the traditional constraint of cache coherency improves write performance by eliminating unneeded cache line fills. This technique conserves backing store bandwidth during many write operations while having negligible impact on the cache's read performance.
1. A method for servicing requests to a backing store using a cache comprising at least one structure, the cache structure being associated with at least one address in the backing store, the method comprising:
(a) receiving a request to write to at least one address in the backing store;
(b) determining the presence in the cache of a cache structure associated with the at least one address;
(c) if no cache structure is associated with the at least one address, adding data to an invalidated cache structure from at least one address in the backing store that includes the at least one address of the request;
(d) servicing the request using the cache structure of step (c); and
(e) storing a value indicating a contiguous subregion of the structure affected by the servicing of the request.
2. The method of claim 1 wherein the cache structures are multi-byte lines.
3. The method of claim 1 wherein the value is stored in a descriptor comprising a pointer to the cache structure and at least one field for storing a value.
4. The method of claim 1 wherein the cache structure of step (c) is the least recently used structure in the cache.
5. The method of claim 1 wherein step (b) comprises:
(b-1) computing an index value from the at least one address of the request; and
(b-2) examining an entry in a lookup table associated with the computed index value to determine the presence in the cache of a cache structure associated wit the at least one address.
6. The method of claim 1 wherein the contents of the cache structure of step (c) are written to the backing store before the addition of the data from the backing store.
7. The method of claim 6 wherein the written contents of the cache structure are a subset of the contents of the cache structure.
8. The method of claim 1 wherein step (e) comprises:
(e-1) accessing a previously stored subregion value;
(e-2) computing a flew subregion value reflecting the sum of the previously stored subregion value and the value specifying the subregion affected by the servicing of the request, the resulting sum specifying a contiguous range; and
(e-3) storing the computed subregion value.
9. The method of claim 1 further comprising writing the contents of the contiguous subregion of the cache structure to the backing store.
10. The method of claim 1 wherein the size of the cache structures is adjusted to improve write performance.
11. A method for servicing requests to a backing store using a cache comprising at least one structure, the cache structure being associated with at least one address in the backing store, the method comprising:
(a) receiving a request to write to at least one address in the backing store;
(b) determining the presence in the cache of a cache structure associated with the at least one address;
(c) if no cache structure is associated with the at least one address, evicting a first structure in the cache;
(d) servicing the request using the first cache structure; and
(e) storing a value indicating a contiguous subregion of the first cache structure affected by the servicing of the request.
12. The method of claim 11 further comprising servicing a second request to read from at least one address in the backing store, wherein the at least one address of the second request is: (i) associated with the first cache structure, and (ii) contained in the addresses associated with the contiguous subregion of the first cache structure.
13. The method of claim 11 further comprising:
receiving a second request to read from at least one address in the backing store, wherein the at least one address of the second request is: (i) associated with the first cache structure, and (ii) not contained in the addresses associated with the contiguous subregion of the first cache structure;
adding data to a second cache structure from at least one address in the backing store that includes the at least one address of the second request; and
copying the data from a contiguous subregion of the first cache structure to the second cache structure.
14. The method of claim 13 further comprising servicing the second request using the second cache structure.
15. The method of claim 13 further comprising replacing the first cache structure with the second cache structure.
16. The method of claim 15 further comprising modifying the stored subregion value to reflect the sum of the previously-stored subregion value for the first cache structure and the value specifying the subregion affected by the servicing of the second request, the resulting sum specifying a contiguous range.
17. The method of claim 11 further comprising:
receiving a second request to write to at least one address in the backing store, the at least one address of the second request is: (i) associated wit the first cache structure, and (ii) not contained in the addresses associated with the contiguous subregion of the first cache structure;
adding data to a second structure in the cache from at least one addresses in the backing store that includes the at least one address of the second request; and
copying the data from the contiguous subregion of the first cache structure to the second cache structure.
18. The method of claim 17 further comprising servicing the second request using the second cache structure.
19. The method of claim 17 further comprising replacing the first cache structure with the second cache structure.
20. The method of claim 19 further comprising modifying the stored subregion value to reflect the sum of the previously-stored subregion value for the first cache structure with the value specifying the subregion affected by the servicing of the second request, the resulting sum specifying a contiguous range.
21. The method of claim 11 wherein the cache structures are multi-byte lines.
22. The method of claim 11 wherein the value is stored in a descriptor comprising a pointer to the cache stricture and at least one field for storing a value.
23. The method of claim 11 wherein the structure of step (c) is the least recently used structure in the cache.
24. The method of claim 11 wherein step (b) comprises:
(b-1) computing an index value from the at least one address of the request; and
(b-2) examining an entry in a lookup table associated with the computed index value to determine the presence in the cache of a cache structure associated wit the at least one address.
25. The method of claim 11 wherein the contents of the cache structure of step (c) are written to a backing store before the first structure is invalidated.
26. The method of claim 25 wherein the written contents of the cache structure are a subset of the contents of the cache structure.
27. The method of claim 11 further comprising writing the contents of the contiguous subregion of the cache structure to the backing store.
28. The method of claim 11 wherein the size of the cache structures is adjusted to improve write performance.