8f6433336d
kernel.sched_migration_cost_ns=5000000 was being defined here even though it was defined in the included latency-performance profile. Removed to eliminate redundancy.
46 lines
1.5 KiB
Text
46 lines
1.5 KiB
Text
#
|
|
# tuned configuration
|
|
#
|
|
|
|
[main]
|
|
summary=Optimize for HPC compute workloads
|
|
description=Configures virtual memory, CPU governors, and network settings for HPC compute workloads.
|
|
include=latency-performance
|
|
|
|
[vm]
|
|
# Most HPC application can take advantage of hugepages. Force them to on.
|
|
transparent_hugepages=always
|
|
|
|
[disk]
|
|
# Increase the readahead value to support large, contiguous, files.
|
|
readahead=>4096
|
|
|
|
[sysctl]
|
|
# Forces hugepages to be allocated on non-hotpluggable memory
|
|
vm.hugepages_treat_as_movable=0
|
|
|
|
# Keep a reasonable amount of memory free to support large mem requests
|
|
vm.min_free_kbytes=135168
|
|
|
|
# Most HPC applications are NUMA aware. Enabling zone reclaim ensures
|
|
# memory is reclaimed and reallocated from local pages. Disabling
|
|
# automatic NUMA balancing prevents unwanted memory unmapping.
|
|
vm.zone_reclaim_mode=1
|
|
kernel.numa_balancing=0
|
|
|
|
# Busy polling helps reduce latency in the network receive path
|
|
# by allowing socket layer code to poll the receive queue of a
|
|
# network device, and disabling network interrupts.
|
|
# busy_read value greater than 0 enables busy polling. Recommended
|
|
# net.core.busy_read value is 50.
|
|
# busy_poll value greater than 0 enables polling globally.
|
|
# Recommended net.core.busy_poll value is 50
|
|
net.core.busy_read=50
|
|
net.core.busy_poll=50
|
|
|
|
# TCP fast open reduces network latency by enabling data exchange
|
|
# during the sender's initial TCP SYN. The value 3 enables fast open
|
|
# on client and server connections.
|
|
net.ipv4.tcp_fastopen=3
|
|
|
|
|