Meta Introduces ‘Transparent Memory Offload’ Feature to Save Massive Memory on Linux Servers
Meta Introduces ‘Transparent Memory Offload’ Feature to Save Massive Memory on Linux Servers.
Engineers at Meta (formerly Facebook) blogged about a new Linux kernel feature called ” Transparent memory offloading “, which they say can save 20% to 32% of memory per Linux server.
Transparent Memory Offload (TMO), Meta‘s solution for heterogeneous data center environments, introduces a new Linux kernel mechanism to measure work lost due to CPU, memory, and I/O resource shortages in real-time.
Guided by this information, TMO automatically adjusts the amount of memory to offload to heterogeneous devices, such as compressed memory or SSDs.
It does this based on the device’s performance characteristics and the app’s susceptibility to slower memory accesses.
The memory savings of TMO can be broken down into application, data center memory tax, and application memory tax.
The graph below shows the relative memory savings achieved by TMO for eight representative applications using different offload backends (compressed memory or SSD).
The blog post emphasizes that the TMO function has been proven for a long time, has high stability, and is very efficient in memory optimization for Linux servers:
TMO has been running in Meta’s production environment for over a year, saving 20% to 32% of total memory across millions of Linux servers in a massive data center fleet .
We have successfully uploaded TMO’s operating system components into the Linux kernel.
Interested friends can check the Meta blog for more technical information about Meta Transparent Memory Offload (TMO).