The　possibilities　of　thread　starvation　and　priority　inversion　can　create　serious　problems　and　make　fairness　a　critical　issue.　When　they　arise　they　can　render　the　Operating　System　(OS)　scheduler　ineffective　if　no　fair　cache　sharing　is　provided　by　the　hardware.　This　paper　proposes　a　dynamic　fair　partitioning　policy　of　shared　cache　in　order　to　improve　the　fairness　of　shared　cache　between　threads　in　a　chip　multiprocessor.　Threads　share　a　cache　while　simulating　that　they　are　running　alone　with　a　dedicated　cache.　At　the　end　of　each　time　interval　T,　a　policy　evaluates　the　metrics　of　the　threads　based　on　shared　and　dedicated　cache,　and　will　then　repartition　the　cache　to　improve　the　fairness.　Experimental　data　shows　that　the　dynamic　fair　partitioning　policy　of　shared　cache　can　effectively　enhance　fairness　and　slightly　increase　throughput　for　chip　multiprocessors　when　compared　to　LRU　or　other　forms　of　partitioning　of　shared　cache.　By　analyzing　the　relationship　between　fairness　and　efficiency,　it　can　be　shown　that　running　different　combinations　of　applications　on　a　system　leads　to　different　changes　in　fairness　and　efficiency.　Therefore,　the　relationship　between　fairness　and　efficiency　is　closely　related　to　the　applications　running　on　the　system.