Skip to content
Tutorials Center
My learning

Topics

Lesson 27 of 37 ยท Analysis and Reporting

Reading the JMeter Statistics Table and JTL Files

Understand JTL result files and the Aggregate Report columns: samples, average, median, percentiles, min, max, error rate and throughput, with a runnable JTL parser.

A JMeter run produces a lot of numbers. Before you can judge a test, you need to know where those numbers come from and what each column means. Everything starts with the JTL file: a log with one line per sample (one request and its response). Listeners, the Aggregate Report and the HTML dashboard are all different views of that file. JMeter cannot run on this site, so this lesson works with a small sample file parsed by a Python stand-in that computes the same statistics.

The JTL file

By default, a non-GUI run with -l results.jtl writes CSV with a header row:

TEXT
timeStamp,elapsed,label,responseCode,responseMessage,threadName,dataType,success,failureMessage,bytes,sentBytes,grpThreads,allThreads,URL,Latency,IdleTime,Connect
1735689600000,182,GET /products,200,OK,Users 1-1,text,true,,5120,310,1,1,https://shop.test/products,120,0,35

The columns you use the most:

Column Meaning
timeStamp start time in milliseconds since 1970
elapsed total time from sending the request to receiving the last byte of the response (ms)
label the sampler's name; this is how results are grouped
responseCode HTTP status, or an exception name such as Non HTTP response code: java.net.SocketTimeoutException
success true or false, after all assertions
bytes, sentBytes response and request size
grpThreads, allThreads active threads in the group and in total at that moment
Latency time to the first byte of the response (includes connect time in JMeter)
Connect time to establish the TCP (and TLS) connection

elapsed is what most people call response time. Latency is "time to first byte". The difference between them is download time; if latency is almost all of elapsed the server is slow to respond, and if latency is small but elapsed is large the payload transfer is slow.

The Aggregate Report / Summary columns

Column Meaning
# Samples number of requests with that label
Average arithmetic mean of elapsed
Median 50th percentile: half of the requests were faster
90% Line, 95% Line, 99% Line percentiles: 90 percent of requests were at least this fast
Min, Max fastest and slowest sample
Error % share of samples with success=false
Throughput completed requests per second (or minute/hour)
Received KB/sec, Sent KB/sec network volume
Std. Dev. spread of response times

A label with a transaction controller is reported as one row whose time is the sum of its children, and with Generate parent sample off you also get the child rows. That is why many teams wrap each business step (such as "Checkout") in a Transaction Controller: it gives a business-level number.

The Total row averages across all labels. A fast login endpoint called many times can hide a slow checkout. Always read per-label rows, not the total.

Parsing a JTL file

This stand-in reads a small embedded JTL, groups by label, and prints the same columns as the Aggregate Report. The percentile uses the nearest-rank method, which is close to what JMeter shows.

Python Stand-in: compute an aggregate report from JTL text

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Notice how the GET /products average is pulled up by two slow samples while its median stays at 150 ms. The next lesson explains why that matters.

Throughput, concurrency and the relationship between them

Three numbers describe load and are linked by Little's Law: concurrent users = throughput x (average response time + think time). If 100 users each wait 4 seconds between requests and the average response is 1 second, throughput is about 100 / 5 = 20 requests per second. If response times double, throughput drops in a closed-loop test, because users spend more time waiting. This is why a decreasing throughput with constant threads often signals a slowdown on the server, not the generator.

Warm-up, ramp-up and the steady state

Do not read numbers from the whole run blindly. Ramp-up (threads starting) and ramp-down distort averages. Compute statistics for a steady-state window: a period when the number of active threads is constant and the system is warm (caches filled, JIT compiled). The dashboard filters can exclude early and late samples (see HTML dashboard and custom reports), and you can cut a JTL file with a small script before analysis.

Common mistakes

  • Reporting the average alone (see percentiles and averages).
  • Mixing different test runs in one file.
  • Forgetting that failed requests are often fast (an immediate 500), which lowers averages while the system is actually broken. Always read response times next to the error rate.
  • Comparing runs with different data or different ramp-ups.
  • Reading the Summary Report from the GUI listener for a long run; use the JTL and the dashboard.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Which JTL column holds the total response time of a sample?

    elapsed is the time from request start to the last byte of the response. Latency is the time to the first byte.
  2. 2. Why can a failed run show a great average response time?

    Fast failures pull the average down, so always read the error rate beside the response times.
  3. 3. What does the 90% Line column mean?

    It is the 90th percentile of response times.
  4. 4. Which part of a run should you analyse?

    Ramp-up and warm-up periods distort results; analyse the steady state.

Summary

  • The JTL file has one row per sample; every report is a view of it.
  • elapsed is the response time; Latency is time to first byte.
  • Aggregate rows give samples, mean, median, percentiles, min, max, error rate and throughput.
  • Read per-label rows, not only the Total.
  • Analyse the steady state and always read errors next to times.

Next: Percentiles and Averages