Intel Sub-NUMA Clustering Will Stop Clashing With Resource Director On Linux 6.11+
([Intel] 5 Hours Ago
SNC + RDT)
- Reference: 0001478788
- News link: https://www.phoronix.com/news/Intel-SNC-RDT-Behaves-Linux-611
- Source link:
For the past year and a half Intel engineers have been working on Linux kernel improvements for [1]Sub-NUMA Clustering (SNC) in the presence of Resource Director Technology (RDT) . Intel has been advising its customers not to use Sub-NUMA Clustering when making use of Resource Director Technology since [2]these features would effectively fight eachother . Well, with the Linux 6.11 kernel that's finally being addressed.
SNC allows partitioning the CPU cores / L3 cache / memory into multiple NUMA domains and can help enhance performance for NUMA-optimized workloads. Intel RDT meanwhile provides for monitoring and greater controls over resources around the last level cache handling and memory bandwidth use. These two features would collide and thus Intel has recommended both aren't engaged concurrently. After [3]many rounds of patch review , SNC + RDT together should behave with code now submitted for the Linux 6.11 kernel.
The [4]x86/cache pull request was submitted from tip/tip.git to the mainline kernel today. As explained in that pull request:
"Enable Sub-NUMA clustering to work with resource control on Intel by teaching resctrl to handle scopes due to the clustering which partitions the L3 cache into sets. Modify and extend the subsystem to handle such scopes properly"
So now those using Intel servers can enjoy both Sub-NUMA Clustering and Resource Director Technology for optimizing performance without the features clashing.
[1] https://www.phoronix.com/news/Linux-resctrl-Sub-NUMA-Cluster
[2] https://www.phoronix.com/news/Intel-SNT-RDT-Fixing-Up
[3] https://www.phoronix.com/news/Intel-SNC-With-RDT-Linux-v19
[4] https://lore.kernel.org/lkml/20240716065458.GAZpYZQhh0PBItpD1k@fat_crate.local/
SNC allows partitioning the CPU cores / L3 cache / memory into multiple NUMA domains and can help enhance performance for NUMA-optimized workloads. Intel RDT meanwhile provides for monitoring and greater controls over resources around the last level cache handling and memory bandwidth use. These two features would collide and thus Intel has recommended both aren't engaged concurrently. After [3]many rounds of patch review , SNC + RDT together should behave with code now submitted for the Linux 6.11 kernel.
The [4]x86/cache pull request was submitted from tip/tip.git to the mainline kernel today. As explained in that pull request:
"Enable Sub-NUMA clustering to work with resource control on Intel by teaching resctrl to handle scopes due to the clustering which partitions the L3 cache into sets. Modify and extend the subsystem to handle such scopes properly"
So now those using Intel servers can enjoy both Sub-NUMA Clustering and Resource Director Technology for optimizing performance without the features clashing.
[1] https://www.phoronix.com/news/Linux-resctrl-Sub-NUMA-Cluster
[2] https://www.phoronix.com/news/Intel-SNT-RDT-Fixing-Up
[3] https://www.phoronix.com/news/Intel-SNC-With-RDT-Linux-v19
[4] https://lore.kernel.org/lkml/20240716065458.GAZpYZQhh0PBItpD1k@fat_crate.local/
npwx