Most popular speedup sysctl options for Proxmox. Put in /etc/sysctl.d/

sysctl-proxmox-tune.conf

###
# Proxmox or other server kernel params cheap tune and secure.
# Try it if you have heavy load on server - network or memory / disk.
# No harm assumed but keep your eyes open.
#
# @updated: 2020-02-06 - more params used, adjust some params values, more comments on params
#

### NETWORK ###

# Timeout broken connections faster (amount of time to wait for FIN)
net.ipv4.tcp_fin_timeout = 10

# Wait a maximum of 5 * 2 = 10 seconds in the TIME_WAIT state after a FIN, to handle
# any remaining packets in the network.
# load module nf_contrack if needed
net.netfilter.nf_conntrack_tcp_timeout_fin_wait = 5

# Keepalive optimizations
# By default, the keepalive routines wait for two hours (7200 secs) before sending the first keepalive probe,
# and then resend it every 75 seconds. If no ACK response is received for 9 consecutive times, the connection is marked as broken.
# The default values are: tcp_keepalive_time = 7200, tcp_keepalive_intvl = 75, tcp_keepalive_probes = 9
# We would decrease the default values for tcp_keepalive_* params as follow:
# Disconnect dead TCP connections after 10 minutes
net.ipv4.tcp_keepalive_time = 600
# Determines the wait time between isAlive interval probes (reduce from 75 sec to 15)
net.ipv4.tcp_keepalive_intvl = 15
# Determines the number of probes before timing out (reduce from 9 sec to 5 sec)
net.ipv4.tcp_keepalive_probes = 5

# allow that much active connections
net.core.somaxconn = 256000

# Protection from SYN flood attack.
net.ipv4.tcp_syncookies = 1
# Only retry creating TCP connections twice
# Minimize the time it takes for a connection attempt to fail
net.ipv4.tcp_syn_retries = 2
net.ipv4.tcp_synack_retries = 2
net.ipv4.tcp_orphan_retries = 2

# Handle SYN floods and large numbers of valid HTTPS connections
net.ipv4.tcp_max_syn_backlog = 40000

# Increase the length of the network device input queue
net.core.netdev_max_backlog = 50000

# Faster full-speed than cubic
# And faster recover if connection looses packets
net.ipv4.tcp_congestion_control = yeah
# http://lwn.net/Articles/616241/
net.core.default_qdisc = fq_codel

# Increase ephermeral IP ports
net.ipv4.ip_local_port_range = 10000    60000

# Broken combined
net.ipv4.tcp_tw_reuse = 0
# The net.ipv4.tcp_tw_recycle has been removed from Linux 4.12 on 2017.
# Removed by upstream kernel, absent since PVE 5.1.
# So comment it out if you use PVE 5.1+
# Let's mark it - PVE3
net.ipv4.tcp_tw_recycle = 0

# Don't need IPv6 for now
# If you use IPv6 - comment this line
net.ipv6.conf.all.disable_ipv6 = 1

# https://www.serveradminblog.com/2011/02/neighbour-table-overflow-sysctl-conf-tunning/
net.ipv4.neigh.default.gc_thresh1 = 1024
net.ipv4.neigh.default.gc_thresh2 = 2048
net.ipv4.neigh.default.gc_thresh3 = 4096

# http://www.opennet.ru/opennews/art.shtml?num=44945
net.ipv4.tcp_challenge_ack_limit = 9999

# Don't slow network - save congestion window after idle
# https://github.com/ton31337/tools/wiki/tcp_slow_start_after_idle---tcp_no_metrics_save-performance
net.ipv4.tcp_slow_start_after_idle = 0

# If we must send packets at first place, but throughput is on second
# Or many small packets.
net.ipv4.tcp_low_latency = 1

#### PVE ####

# Allow a high number of timewait sockets
net.ipv4.tcp_max_tw_buckets = 2000000

# PVE 3
net.ipv4.tcp_max_tw_buckets_ub = 65000


# Increase Linux autotuning TCP buffer limits
net.ipv4.tcp_wmem = 4096        65536   16777216
net.ipv4.tcp_rmem = 4096        87380   16777216
net.core.rmem_max = 16777216
net.core.wmem_max = 16777216
net.core.optmem_max = 65536


# If your servers talk UDP, also up these limits
net.ipv4.udp_rmem_min = 8192
net.ipv4.udp_wmem_min = 8192

# Sockets/UDP query length
net.unix.max_dgram_qlen = 1024

# http://vds-admin.ru/unix-linux/oshibki-v-dmesg-vida-nfconntrack-table-full-dropping-packet
# load module nf_contrack if needed
net.netfilter.nf_conntrack_max = 1048576
net.nf_conntrack_max = 1048576


### MEMORY ###

# do less swap but not disable it
vm.swappiness = 2

# allow application request allocation of virtual memory
# more than real RAM size (or OpenVZ/LXC limits)
vm.overcommit_memory = 1

# https://major.io/2008/12/03/reducing-inode-and-dentry-caches-to-keep-oom-killer-at-bay/
vm.vfs_cache_pressure = 500

# time in  centi-sec. i.e. 100 points = 1 second
# delayed write of dirty data
vm.dirty_writeback_centisecs = 3000
# flush from memory old dirty data
vm.dirty_expire_centisecs = 18000

##
# Adjust vfs cache
# https://lonesysadmin.net/2013/12/22/better-linux-disk-caching-performance-vm-dirty_ratio/
# Decriase dirty cache to faster flush on disk
vm.dirty_background_ratio = 5
vm.dirty_ratio = 10

#### PVE 3 ####

# Only on Proxmox 3.x with OpenVZ
ubc.dirty_ratio = 20
ubc.dirty_background_ratio = 10

# Isolate page cache for VPS.
ubc.pagecache_isolation = 1

### FileSystem ###

##
# Fix: Failed to allocate directory watch: Too many open files
#      in Proxmox 5 + LXC
#      And VM with Bitrix
#      == alot of files

fs.inotify.max_user_instances = 16777216
fs.inotify.max_queued_events = 32000
fs.inotify.max_user_watches = 64000


### Security ###

# http://www.opennet.ru/opennews/art.shtml?num=47792
kernel.unprivileged_bpf_disabled=1

# http://www.opennet.ru/opennews/art.shtml?num=49135
# https://community.rti.com/kb/how-can-i-improve-my-throughput-performance-linux
net.ipv4.ipfrag_high_thresh=8388608
net.ipv4.ipfrag_low_thresh=196608
net.ipv6.ip6frag_high_thresh=8388608
net.ipv6.ip6frag_low_thresh=196608

# http://www.opennet.ru/opennews/art.shtml?num=50889
net.ipv4.tcp_sack = 0
net.ipv4.tcp_mtu_probing = 0

# Prevent TIME_WAIT attak.
net.ipv4.tcp_rfc1337 = 1


### OTHER ###

### PVE 3 - 6 kernels ###
# https://tweaked.io/guide/kernel/
# Don't migrate processes between CPU cores too often
# kernel <= 5.4 (ie Proxmox 6)
kernel.sched_migration_cost_ns = 5000000
# Kernel >= 2.6.38 (ie Proxmox 4+)
kernel.sched_autogroup_enabled = 0

vm.vfs_cache_pressure controls how many % of memory used for cache of inodes. So it sets up as many as possible.
vm.dirty_expire_centisecs controls in 10 secs how long data stays in memory before it will be written on disk.
vm.dirty_writeback_centisecs controls in 10 secs how often background checks will find out which data is expired.

More here: https://www.baeldung.com/linux/file-system-caching

tcp_low_latency (Boolean; default: disabled; since Linux 2.4.21/2.6; obsolete since Linux 4.14)

Since Linux 4.14, this file still exists, but its value is ignored.

https://man7.org/linux/man-pages/man7/tcp.7.html