Feature/optimize cache - Githubissues

LLNL / metall

Persistent memory allocator for data-centric analytics

Other

53 stars 13 forks source link

Feature/optimize cache #295

Closed KIwabuchi closed 1 year ago

KIwabuchi commented 1 year ago

This PR is for a completely new object cache implementation.

The new cache is:

Uses much less memory
Significantly faster initialization time, especially on a machine that has many CPU cores
More compact and packed internal data structure
Yet, achieves the same Metall performance on the dynamic graph construction workload

Simple create and close test results (just creating and closing Metall manager).

New ver Create 0.0440739s Close 0.00165344s
Old ver Open 0.658001s Close 0.116477s