Method and apparatus for optimal cache sizing and configuration for large memory systems

1. A configurable memory subsystem of a computing system, the configurable memory subsystem comprising:

a memory;

a configurable memory subsystem tool to:

express performance metrics of the computing system hosting applications that utilize the memory as an explicit function of configuration parameters of the memory and workload parameters of the memory by:

collecting histogram data for system performance statistics of the computing system for a plurality of different arrangements of the configuration parameters;

using the histogram data to establish a parametric density function comprising a plurality of density parameters and the plurality of arrangements of the configuration parameters;

using the parametric density function to determine a density function of the performance metrics as a function of the configuration parameters; and

using the density function to determine the explicit function of the performance metrics, configuration parameters and workload parameters;

identify a performance goal for the computing system, the performance goal comprising values for the performance metrics; and

use the values for the performance metrics in the explicit function of performance metrics, configuration parameters and workload parameters to calculate values for the configuration parameters that achieve the identified performance goal; and

a configuration controller in communication with the memory and the configurable memory subsystem tool to implement the calculated values of the configuration parameters in the memory.

2. The configurable memory subsystem of claim 1 , wherein the memory comprises a capacity of at least 10 gigabytes.

3. The configurable memory subsystem of claim 1 , wherein the memory comprises a hybrid memory system.

4. The configurable memory subsystem of claim 3 , wherein the hybrid memory system comprises two distinct types of memory selected from the group consisting of static random access memory, dynamic random access memory, phase change memory, NAND flash memory and NOR flash memory.

5. The configurable memory subsystem of claim 1 , wherein the memory comprises a cache, and the workload parameters comprise load on the cache, temporal correlation within the cache, spatial correlation within the cache or combinations thereof.

6. The configurable memory subsystem of claim 1 , wherein the configuration parameters comprise hardware parameters, software parameters or combinations thereof.

7. The configurable memory subsystem of claim 6 , wherein the memory comprises a cache and a main memory portion separate from the cache, and the hardware parameters comprise cache size, total memory size, line size within the cache, block size within the main memory portion, associativity within the cache or combinations thereof.

8. The configurable memory subsystem of claim 6 , wherein the software parameter comprises a replacement algorithm.

9. The configurable memory subsystem of claim 1 , wherein the configurable memory subsystem tool further comprises an optimization engine to solve the explicit function of performance metrics, configuration parameters and workload parameters.

10. The configurable memory subsystem of claim 1 , wherein the configurable memory subsystem tool further comprises a memory performance engine to connect system performance statistics to the system performance metrics and to use this connection to develop the explicit function.

11. The configurable memory subsystem of claim 10 , wherein the system performance metrics comprise a miss ratio comprising a ratio of misses to a memory cache to hits to the memory cache.

12. The configurable memory subsystem of claim 10 , wherein the memory comprises a cache and the system performance statistics comprise cache performance statistics.

13. The configurable memory subsystem of claim 12 , wherein the cache performance statistics comprise cache residency time, single residency time, inter-hit time or combinations thereof.

14. The configurable memory subsystem of claim 1 , wherein the configurable memory subsystem tool is further configured to:

process the histogram data associated with each arrangement of the configuration parameters to determine an empirical density for each arrangement of the configuration parameters, the empirical density comprising a probability that a value of a given system performance statistic falls between a given set of values; and

use the histogram data to establish the parametric density function further comprising using the empirical density for each arrangement of the configuration parameters to establish the parametric density function.

15. The configurable memory subsystem of claim 14 , wherein the histogram data is collected from the computing system during runtime.

16. The configurable memory subsystem of claim 14 , wherein the histogram data is collected from a memory cache simulation of the computing system.

17. The configurable memory subsystem of claim 14 , wherein the histogram data represent different time ranges and varying granularities.

18. The configurable memory subsystem of claim 14 , wherein configurable memory subsystem tool:

uses the parametric density function for the different arrangements of the configuration parameters to establish a functional dependence between the density parameters and the configuration parameters; and

uses the established functional dependence to determine the density function.

19. The configurable memory subsystem of claim 18 , wherein configurable memory subsystem tool uses the explicit function to determine mean, standard deviation and higher order moments of the performance metrics in terms of the configuration parameters.

20. The configurable memory subsystem of claim 19 , wherein the configurable memory subsystem tool uses the mean, standard deviation and higher order moments to identify a preferred set of configuration parameters for the identified performance goal.