Prefetch system and method
AssigneeExablade Corporation
InventorTed Bagg
Life of this patent
Estimating the term…
Abstract
A system and method for prefetching data. Address logs are separated into streams and a model associated with each stream. Each stream address is forecasted according to its respective model and pages corresponding to one or more forecasted stream addresses are retrieved from memory based on their respective models.
Claims20
A method for prefetching data, comprising:
- separating address logs into streams;
- modeling each stream;
- forecasting each stream address according to its respective model;
- and
- retrieving pages corresponding to one or more forecasted stream addresses based on their respective models.
An article comprising a nontransitory computer-readable medium having instructions thereon, wherein the instructions, when executed in a computer, create a system for executing a method of soliciting predictions, the method comprising:
- separating address logs into streams;
- modeling each stream;
- forecasting each stream address according to its respective model;
- and
- retrieving pages corresponding to one or more forecasted stream addresses based on their respective models.
A computer system, comprising:
- a cache;
- and
- a plurality of processors connected to the cache, wherein the cache includes a prefetch mechanism, wherein the prefetch mechanism:
- separates address logs into streams;
- models each stream;
- forecasts each stream address according to its respective model;
- and
- retrieves pages corresponding to one or more forecasted stream addresses based on their respective models.
An asynchronous genetic method, comprising:
- defining a fixed population of competitive slots;
- assigning no more than one individual to each slot;
- progressing the slots along a finite set of states and queues, whose service routines determine each slot's subsequent state, wherein progressing includes performing a fitness evaluation of the individual in one or more of the slots.
A method of deciding pages to replace in a cache, comprising:
- storing addresses of pages of interest in a graded partial order least recently used (LRU) list;
- and
- modifying the list to track age of pages in the cache.
A method of reproduction of a genetic algorithm used to train a neural network, the method comprising:
- maintaining fitness ranks for individuals;
- determining the probability of each possible reproductive combination according to sum and difference of fitness ranks.
Dependencies are inferred from the claim text.