Meta engineer Gregory Price and his team have introduced CRAM, a new Linux memory compression method presented at the Linux Plumbers' Conference. Unlike ZRAM, CRAM keeps data in memory using a private NUMA node instead of treating it as a block device swap.
CRAM supports standard Linux mechanisms like migration and ballooning while offering massive speed gains. In read-only tests, it reached 489 million operations per second, outperforming ZRAM by 452x. Even with 20% writes, it remains 5.4x faster than ZRAM.
The technology uses a Chicken Bit to manage allocations during busy periods, though writes require folio migration. While CRAM lacks a way to determine exact logical RAM capacity due to data compressibility, it provides a high-speed alternative for memory.