Reducing page faults in host OS following a live partition mobility event

摘要

Page faults during partition migration from a source computing system to a destination computing system are reduced by assigning each page used by a process as being hot or cold according to their frequency of use by the process. During a live partition migration, the cold or coldest (least frequently used) pages are copied to the destination server first, followed copying the warmer (less frequently used) and concluded by copying the hottest (most frequently used) pages. After all dirtied pages have been refreshed, cutover from the instance on the source server to the destination server is made. By transferring the warm and hot pages last (or later) in the migration process, the number of dirtied pages is reduced, thereby reducing page faults subsequent to the cutover.

主权项

1. A method for reducing page faults during partition migration from a source computing system to a destination computing system comprising:
designating in a computer readable tangible storage memory device a frequency of use designation for each of a plurality of memory pages being used by a first instance of a computer process currently being executed on a source computing system, wherein the source computing system has a microprocessor; responsive to a live partition migration event beginning, rank ordering the plurality of memory pages according to the frequency of use designations to create a sequence of copying during the migration event, wherein the rank order of each partition can take the state of hot and cold coupled with a type of access designation of read-only and read-write; copying read-write pages from the source computing system to a destination computing system according to a sequence of cold to hot; copying read-only pages from the source computing system to the destination computing system according to a sequence of hot to cold; refreshing one or more partition copies on the destination computing system in which any pages which have been dirtied on the source computing system since the beginning of the migration event; and responsive to all pages being copied and refreshed as applicable, transferring execution of the first instance of the computer process to a second instance of the computer process executing on the destination computing system, and redirecting workflow from the source computing system to the destination computing system associated with the computer process.