How to Increase somaxconn Value for Better Network Performance in Linux


2 views

The somaxconn parameter in Linux determines the maximum number of backlog connections a socket can hold when the system is not accepting new connections. The default value is typically 128, which might be insufficient for high-traffic servers.

# Check current value
cat /proc/sys/net/core/somaxconn

You can temporarily modify the value using sysctl (effective until reboot):

# Set to 1024 temporarily
echo 1024 > /proc/sys/net/core/somaxconn

# Or using sysctl
sudo sysctl -w net.core.somaxconn=1024

For persistent changes across reboots:

# Edit sysctl.conf
sudo nano /etc/sysctl.conf

# Add this line
net.core.somaxconn = 1024

# Apply changes
sudo sysctl -p

After modification, verify the change:

sysctl net.core.somaxconn
cat /proc/sys/net/core/somaxconn

Note that some applications (like Nginx, Redis) have their own backlog settings that may need adjustment to match or exceed somaxconn:

# For Redis example
# In redis.conf:
tcp-backlog 511
  • Higher values consume more memory
  • Needs root privileges for permanent changes
  • Should match or be lower than application-specific backlog settings
  • Monitor system performance after changes

For web servers, you might test with different values:

# Apache Benchmark test
ab -n 10000 -c 1000 http://yourserver/

Compare results before and after adjusting somaxconn to measure improvement.

The somaxconn parameter defines the maximum number of backlog connections a socket can queue when handling new connection requests. You can view the current value using:

cat /proc/sys/net/core/somaxconn

Typical default values range from 128 to 1024, which might be insufficient for high-traffic servers.

For immediate changes (lost on reboot):

echo 1024 > /proc/sys/net/core/somaxconn
# Or with sysctl:
sudo sysctl -w net.core.somaxconn=1024

For persistent configuration, add this line to /etc/sysctl.conf:

net.core.somaxconn = 1024

Then apply with:

sudo sysctl -p

Many server applications have their own backlog settings that must align with somaxconn. For example, in Nginx:

server {
    listen 80 backlog=1024;
    ...
}

For Python socket servers:

import socket

s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind(('0.0.0.0', 8080))
s.listen(1024)  # Should be <= somaxconn value

Check current TCP connection queue status:

ss -lnt

Output columns show Send-Q (backlog limit) and Recv-Q (current queue length). Monitor with:

watch -n 1 'ss -lnt'
  • If changing somaxconn doesn't take effect, check application-specific limits
  • Kernel versions before 2.4.25 had a hardcoded maximum of 128
  • Values beyond 65535 are generally unnecessary and may waste memory
  • Consider net.ipv4.tcp_max_syn_backlog for SYN flood protection