Mastering Tomcat Performance: Tuning Threads, Connections, and Linux Kernel Settings

Tomcat Self‑Optimization

Tomcat tuning focuses on several server.xml attributes. Understanding each parameter’s purpose is essential.

1. maxThreads

Defines the maximum number of request‑processing threads. The default is 200. For CPU‑bound workloads, a smaller value reduces thread‑switching overhead; for I/O‑bound workloads, a larger value increases concurrency. Setting it too high (e.g., 3000) can cause response times to spike because the CPU spends most of its time switching threads.

2. acceptCount

Specifies the maximum number of queued requests when all maxThreads are busy. Default is 100. If the queue is full, new connections are refused. The effective queue length is the smaller of acceptCount and the Linux kernel parameter net.core.somaxconn (default 128).

Experiment:

Set acceptCount=3000 and net.core.somaxconn=8192; ss -lt shows send_q=3000.

Set acceptCount=10000 with the same net.core.somaxconn; send_q=8192, confirming the kernel limit dominates.

3. maxConnections

Maximum simultaneous connections the server will accept. When reached, the server accepts one extra connection but does not process it until the count drops below maxConnections. Default varies: NIO/NIO2 → 10000, APR/native → 8192. The value can be disabled with -1.

4. connectionTimeout

Time (ms) Tomcat waits for the request URI after a socket is accepted. Default in Tomcat’s server.xml is 20000 ms (20 s). Setting -1 disables the timeout.

Typical server.xml modifications:

vi server.xml
# Replace the commented Executor block
<Executor name="tomcatThreadPool" namePrefix="catalina-exec-" maxThreads="1500" minSpareThreads="50" prestartminSpareThreads="true"/>
# Update the Connector
<Connector executor="tomcatThreadPool" port="8009" protocol="org.apache.coyote.http11.Http11Nio2Protocol" connectionTimeout="20000" maxConnections="10000" redirectPort="8443" acceptCount="1500"/>

Linux Kernel Parameter Optimization

Linux limits affect Tomcat’s ability to handle many concurrent connections.

1. File descriptor limits

Check the per‑user limit with ulimit -u (default 1024). Increase both soft and hard limits in /etc/security/limits.conf:

prouser soft nofile 65536
prouser hard nofile 65536
prouser soft nproc 65536
prouser hard nproc 65536

2. TCP‑related kernel settings

Edit /etc/sysctl.conf and add:

net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_tw_recycle = 1
net.ipv4.tcp_fin_timeout = 30
net.ipv4.ip_local_port_range = 10000 65000
net.ipv4.tcp_max_syn_backlog = 8192
net.ipv4.tcp_max_tw_buckets = 10000
net.core.somaxconn = 8192

Apply changes with sudo /sbin/sysctl -p.

JVM Tuning for Tomcat

Typical JVM options for a high‑performance Tomcat instance:

JAVA_OPTS="$JAVA_OPTS -server -Xmn2000m -Xms4000m -Xmx4000m -XX:PermSize=128m -XX:+UseConcMarkSweepGC -XX:MaxPermSize=512m -Djuli-logback.configurationFile=file:$CATALINA_HOME/conf/logback.xml"

Executor configuration example:

<Executor name="tomcatThreadPool" namePrefix="catalina-exec-" maxThreads="500" minSpareThreads="30" maxIdleTime="60000" prestartminSpareThreads="true" maxQueueSize="100"/>

Key parameters:

maxThreads : recommended 500‑800 based on hardware and workload.

minSpareThreads : initial thread count, default 25.

maxIdleTime : idle thread lifespan in ms (default 60000).

prestartminSpareThreads : pre‑creates the minimum spare threads at startup.

maxQueueSize : maximum waiting queue length before rejecting requests.

Combining proper server.xml settings, Linux kernel tuning, and JVM options yields a balanced Tomcat deployment capable of handling high concurrency with stable response times.