Why Asynchronous Logging Doesn’t Boost TPS – Real Test Results with Logback and Log4j2

Earlier I published an article titled “Logback configuration file improves TPS by 10×”, which attracted many comments and suggestions.

logback performance test shows little difference between synchronous and asynchronous TPS.

Even in 2022 we are still using logback.

After reviewing the feedback, I conducted a new round of testing.

Server Hardware

CPU: 6 cores

Memory: 8 GB

Testing Tools

JMeter

JProfile

APM (New Relic)

Issues with the Previous Test

Sample size was too small (100 threads, 1 loop, total 100 samples).

Only one test run, leading to randomness.

Inconsistent sample counts between runs caused pollution.

In the new test I increased the thread count to 200, loop count to 100, resulting in 20 000 samples per run, and repeated the test five times.

The new results show that TPS differences between synchronous and asynchronous logging are negligible.

Even though asynchronous logging does not significantly increase TPS, it reduces the time spent writing logs.

Using APM to monitor JMeter requests (100 threads, 0 ramp‑up, 100 loops) gave the following averages over five minutes:

Async logging average TPS: 378 rpm.

Sync logging average TPS: 333 rpm, confirming the difference is still small.

Why the change is small: TPS includes three stages – client request, server internal processing, and response. Logging accounts for only about 5.3 % of the total processing time, so optimizing it has limited impact on overall TPS, though it does cut log‑write latency.

Recommendation: Use Async Logging

For Logback, enable asynchronous logging to reduce log latency.

Why Switch to Log4j2

Log4j2 is the successor of Log4j 1.x, incorporates many of Logback’s good designs, and adds major performance improvements. Its asynchronous implementations include Async Append (similar to Logback) and Async Log, which is built on the LMAX Disruptor library for high‑throughput logging.

In the test, Log4j2’s async implementation (Async Log) used the same parameters as the Logback test: 200 threads, 100 loops, five repetitions.

TPS increased by six times!

Log4j2’s async logging reduces log‑write time to near‑zero, and official benchmarks confirm its superiority over Logback.

Official test report: https://logging.apache.org/log4j/log4j-2.3/manual/async.html

Conclusion

If you continue using the Logback framework, enable asynchronous logging.

When choosing a logging framework, prefer Log4j2 for its superior async performance.