Leveraging low-latency memory access

A method of evaluating a query, the method comprising:

• receiving the query at a first instance of a program that compares queries with an index, said first instance running on a first processor of a machine;
• receiving the query at a second instance of said program, said second instance running on a second processor of said machine;
• using said first instance to look up keywords of the query in a first chunk of said index that is stored in a first memory;
• using said second instance to look up keywords of the query in a second chunk of said index that is stored in a second memory, accesses of said first memory by said first processor and accesses of said second memory by said second processor having a lower latency time than accesses of said first memory by said second processor or accesses of said second memory by said first processor;
• and
• returning a result, said result being based on comparison of the query with said first chunk and with said second chunk.

A method of preparing a machine to perform a task, the method comprising:

• dividing data involved in performing the task into a number of partitions;
• creating said number of silos on one processor of said machine, said processor having a plurality of cores, there being at least as many cores as said first number;
• assigning, to each of said silos, an item of software that performs a computational unit of the task and one said partitions;
• and
• executing each item of software, within its respective silo, to operate on the partition in that item of software's silo,
• said task comprising comparing a query to an index, said data comprising said index, each of said partitions comprising a slice of said index, each item of software comprising an instance of a search program that compares said query to a slice of the index that is assigned to the silo in which that instance executes, slices of said index comprising overlapping keywords such that the slice assigned to a first one of the silos comprises a word, the slice assigned to a second one of the partitions comprises said word, and the slice assigned to the first partition associates said word with a different set of documents than the slice assigned to the second partition.