The Linux-ek kernel and modules with ECHO CPU scheduler by Hamad Al Marri and Con Kolivas' hrtimer patches and with some other stuff. Based on Linux-ck maintained by graysky.
- ck-hrtimer - Con Kolivas' hrtimer from -ck.
- ECHO CPU Scheduler - Enhanced CPU Handling Orchestrator, implemented on top of Baby scheduler.
- kernel_compiler_patch enables compiler optimizations for additional CPUs.
- BBR v3 is a congestion control algorithm proposed by Google.
- clearlinux from Intel's Clear Linux project. Provides performance and security optimizations.
- uksmd - Userspace KSM helper daemon.
- le9uo protects the working set under memory pressure to prevent thrashing, avoid high latency and prevent livelock in near-OOM conditions.
- CJKTTY supports displaying CJK Unified Ideographs on Linux tty.
Open /etc/pacman.conf and comment out the original Architecture, then add the new Architecture.
#Architecture = auto
Architecture = x86_64 x86_64_v3
Selecting the correct CPU optimized package.
/lib/ld-linux-x86-64.so.2 --help | grep supported
If x86-64-v3 (supported, searched) is in the output, use the Generic-x86-64-v3 (GENERIC_CPU3).
You should compile it yourself and choose the optimization option that suits you. (NOTE: The package on the releases page is a very small kernel (module list) compiled from the configuration generated by modprobed-db on my intel device, and may only be applicable to intel devices(Coffee Lake), not universal.)
git clone https://github.com/RiverOnVenus/linux-ek.git
cd linux-ek/linux-ek
updpkgsums && makepkg -srci
When you use a kernel compiled by CLANG/LLVM/LTO, some modules that use DKMS need to be recompiled with CLANG/LLVM. Otherwise DKMS will fail.
You need to modify the /etc/dkms/framework.conf file, add two lines to the end of the file: export LLVM=1, export CC=clang.
If you have done that, just reinstall or install the kernel compiled with CLANG/LLVM/LTO and DKMS will not fail again.
Dkms(v3.0.2) support for Clang, use the latest version of dkms and you'll be fine.
This start-up message should appear in the kernel ring buffer when TT in enabled, use:
# dmesg | grep -i 'ECHO CPU'
You can see: ECHO CPU scheduler v6.8 by Hamad Al Marri.
- The default HZ for ECHO is 625HZ - ticks every 1.6ms. No need to increase it since the HighRes clock handles the task preemption in 35us max.
kernel.sched_bs_shared_quotaby default is 35000 (35us) can be tuned with sysctl ex.sysctl kernel.sched_bs_shared_quota=4800000larger values saves CPU caches but reduces interactivity and multitasking.- There are kernel configurations that must be disabled:
- CONFIG_FAIR_GROUP_SCHED
- CONFIG_SCHED_AUTOGROUP
- CONFIG_SCHED_CORE
The special version of uksmd for CachyOS also includes uksmdstats .
➜ uksmdstats
======================================================
uksmd with uksmdstats support. Powered by CachyOS Team
======================================================
Full scans 466
Interval 20 ms
Max page sharing ratio 256
Pages to scan 100
Pages over ratio 3
Duplicated pages 115
Use zero pages 0
Sharing/shared ratio 15.4498
Unshared/sharing ratio 2.58549
Pages sharing 139.953 MiB
Pages shared 9.05859 MiB
Pages unshared 361.957 MiB
General profit 124.181 MiB
# See https://wiki.archlinux.org/title/Sysctl for more information.
# Networking
# Increasing this value for high speed cards may help prevent losing packets:
net.core.netdev_max_backlog = 16384
# The upper limit on how many connections the kernel will accept (default 128):
net.core.somaxconn = 8192
# The default the Linux network stack is not configured for high speed large file transfer across WAN links (i.e. handle more network packets) and setting the correct values may save memory resources:
net.core.rmem_default = 1048576
net.core.rmem_max = 16777216
net.core.wmem_default = 1048576
net.core.wmem_max = 16777216
net.core.optmem_max = 65536
net.ipv4.tcp_rmem = 4096 1048576 2097152
net.ipv4.tcp_wmem = 4096 65536 16777216
# It is also possible increase the default 4096 UDP limits:
net.ipv4.udp_rmem_min = 8192
net.ipv4.udp_wmem_min = 8192
# Enable TCP Fast Open
net.ipv4.tcp_fastopen = 3
# Tweak the pending connection handling
net.ipv4.tcp_max_syn_backlog = 8192
net.ipv4.tcp_max_tw_buckets = 2000000
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_fin_timeout = 10
net.ipv4.tcp_slow_start_after_idle = 0
# Enable MTU probing
net.ipv4.tcp_mtu_probing = 1
# Change TCP keepalive parameters
net.ipv4.tcp_keepalive_time = 120
net.ipv4.tcp_keepalive_intvl = 10
net.ipv4.tcp_keepalive_probes = 6
# Enable CAKE
net.core.default_qdisc = cake
# Virtual memory
# vm.dirty_background_ratio default 10
# vm.dirty_ratio default 20
# For 16GB RAM
vm.dirty_background_ratio = 2
vm.dirty_ratio = 4
# And
# For 8GB RAM
# vm.dirty_background_ratio = 4
# vm.dirty_ratio = 8
# VFS cache
# Decreasing the virtual file system (VFS) cache parameter value
# may improve system responsiveness (default 100)
# vm.vfs_cache_pressure = 50
# Swappiness (default 60)
# For Solid State Drives
vm.swappiness = 100
# See https://chrisdown.name/2018/01/02/in-defence-of-swap.html
#
# Or
#
# Since zram behaves differently than disk swap,
# we can configure the system's swap to take full potential of the zram advantages:
# vm.swappiness = 180
# vm.watermark_boost_factor = 0
# vm.watermark_scale_factor = 125
# vm.page-cluster = 0
# See https://wiki.archlinux.org/title/Zram#Optimizing_swap_on_zram
# cat /sys/block/sda/queue/scheduler
To change the active I/O scheduler to bfq for device sda, use:
# echo bfq > /sys/block/sda/queue/scheduler
Or create file /etc/udev/rules.d/60-ioschedulers.rules:
# set scheduler for NVMe
ACTION=="add|change", KERNEL=="nvme[0-9]n[0-9]", ATTR{queue/scheduler}="bfq"
# set scheduler for SSD and eMMC
ACTION=="add|change", KERNEL=="sd[a-z]|mmcblk[0-9]*", ATTR{queue/rotational}=="0", ATTR{queue/scheduler}="bfq"
# set scheduler for rotating disks
ACTION=="add|change", KERNEL=="sd[a-z]", ATTR{queue/rotational}=="1", ATTR{queue/scheduler}="bfq"
Reboot or force udev#Loading new rules:
If rules fail to reload automatically, use:
# udevadm control --reload
To manually force udev to trigger your rules, use:
# udevadm trigger