Refinement with age As data age they are progressively refined into more compact (lossy) representations.

Suppose, e.g., that data might be stored in three buffers: raw, reduced resolution, histogram. The value per bit stored increases as the data are compressed, yet the value of the data prior to compression is generally higher than that after compression. Given fixed total memory, one optimization is to maximize the expected value by partitioning the data among these classes based upon their values and request probabilities . To simplify, suppose the compression ratios at each step are 100:1 and the histogram must be stored permanently as it is frequently accessed. Thus the optimization reduces to partitioning the memory between the raw and reduced resolution data. Suppose that a raw data record has probability of 0.1 of being requested within a period T of first being acquired and 0.01 otherwise. The revenue per bit is 1 unit. Corresponding figures for the reduced resolution records are twice as large. How should the memory be partitioned to maximize the expected value?