Command-line DNS benchmark tool built to stress test and measure the performance of DNS servers with commodity hardware. This tool typically consumes ~30kB per concurrent connection and can maintain ~30,000 QPS per modern core if your server, OS and network allows you to reach suitable levels of concurrency.
DNStrace bypasses OS resolvers and is provided as a Docker packaged prebuilt static binary.
Basic latency measurement, result checking and histograms are supported.
Currently, only A, AAAA and TXT questions are supported.
$ dnstrace --help
usage: dnstrace [<flags>] <queries>...
A DNS benchmark.
Flags:
--help Show context-sensitive help (also try --help-long
and --help-man).
-s, --server="127.0.0.1" DNS server IP:port to test.
-t, --type=A Query type.
-n, --number=1 Number of queries to issue. Note that the total
number of queries issued =
number*concurrency*len(queries).
-c, --concurrency=1 Number of concurrent queries to issue.
-l, --rate-limit=0 Apply a global questions / second rate limit.
-e, --expect=EXPECT ... Expect a specific response.
-r, --recurse Allow DNS recursion.
--edns0=0 Enable EDNS0 with specified size.
--tcp Use TCP fot DNS requests.
--write=1s DNS write timeout.
--read=4s DNS read timeout.
--codes Enable counting DNS return codes.
--min=400µs Minimum value for timing histogram.
--max=4s Maximum value for histogram.
--precision=[1-5] Significant figure for histogram precision.
--distribution Display distribution histogram of timings to
stdout.
--csv=/path/to/file.csv Export distribution to CSV.
--io-errors Log I/O errors to stderr.
--silent Disable stdout.
--color ANSI Color output.
--version Show application version.
Args:
<queries> Queries to issue.
While dnstrace is helpful for testing round trip latency via public networks,
the code was primarily created to provide an apachebench
style tool for testing your own infrastructure.
It is thus very easy to create significant DNS load with non default settings. Do not do this to public DNS services. You will most likely flag your IP.
go get github.com/redsift/dnstrace will install the binary in your $GOPATH/bin.
On OS-X, the native binary will outperform the Docker container below running under HyperKit significantly e.g. 30% more throughput, 30% lower latency and a 4x decrease in timing spread
This tool is available in a prebuilt image.
docker run redsift/dnstrace --help
If your local test setup lets you reach 50k QPS and above, you can expect the docker networking to add ~2% overhead to throughput and ~8% to mean latency (tested on Linux Docker 1.12.3).
If this is significant for your purposes you may wish to run with --net=host
./dnstrace --completion-script-bash and ./dnstrace --completion-script-zsh will create shell completion scripts.
e.g.
$ eval "$(./dnstrace --completion-script-zsh)"
$ ./dnstrace --concurrency
--codes --distribution --io-errors --precision --server --version
--color --edns0 --max --rate-limit --silent --write
--concurrency --expect --min --read --tcp
--csv --help --number --recurse --type
As you approach thousands of concurrent connections on OS-X, you may run into connection errors due to insufficient file handles or threads. This is likely due to process limits so remember to adjust these limits if you intent to increase concurrency levels beyond 1000.
Note that using sudo ulimit will create a root shell, adjusts its limits, and then exit causing no real effect. Instead use launchctl first on OS-X.
$ sudo launchctl limit maxfiles 1000000 1000000
$ ulimit -n 12288
For long runs, the user can send a SIGHUP via kill -1 pid to get the current progress counts.
$ docker run redsift/dnstrace -n 10 -c 10 --server 8.8.8.8 --recurse redsift.io
Benchmarking 8.8.8.8:53 via udp with 10 conncurrent requests
Total requests: 100 of 100 (100.0%)
DNS success codes: 100
DNS response codes
NOERROR: 100
Time taken for tests: 107.091332ms
Questions per second: 933.8
DNS timings, 100 datapoints
min: 3.145728ms
mean: 9.484369ms
[+/-sd]: 5.527339ms
max: 27.262975ms
DNS distribution, 100 datapoints
LATENCY | | COUNT
+-------------+---------------------------------------------+-------+
3.276799ms | ▄▄▄▄▄▄▄▄ | 2
3.538943ms | ▄▄▄▄▄▄▄▄▄▄▄▄ | 3
3.801087ms | ▄▄▄▄▄▄▄▄▄▄▄▄ | 3
4.063231ms | ▄▄▄▄▄▄▄▄▄▄▄▄ | 3
4.325375ms | ▄▄▄▄▄▄▄▄ | 2
4.587519ms | | 0
4.849663ms | | 0
5.111807ms | ▄▄▄▄ | 1
5.373951ms | ▄▄▄▄ | 1
5.636095ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 4
5.898239ms | ▄▄▄▄▄▄▄▄▄▄▄▄ | 3
6.160383ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 5
6.422527ms | ▄▄▄▄▄▄▄▄▄▄▄▄ | 3
6.684671ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 5
6.946815ms | ▄▄▄▄▄▄▄▄ | 2
7.208959ms | ▄▄▄▄ | 1
7.471103ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 9
7.733247ms | ▄▄▄▄▄▄▄▄ | 2
7.995391ms | ▄▄▄▄▄▄▄▄ | 2
8.257535ms | ▄▄▄▄▄▄▄▄▄▄▄▄ | 3
8.650751ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 5
9.175039ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 11
9.699327ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 6
10.223615ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 4
10.747903ms | ▄▄▄▄ | 1
11.272191ms | ▄▄▄▄ | 1
11.796479ms | | 0
12.320767ms | | 0
12.845055ms | | 0
13.369343ms | | 0
13.893631ms | ▄▄▄▄ | 1
14.417919ms | ▄▄▄▄ | 1
14.942207ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 5
15.466495ms | | 0
15.990783ms | ▄▄▄▄ | 1
16.515071ms | | 0
17.301503ms | | 0
18.350079ms | | 0
19.398655ms | ▄▄▄▄ | 1
20.447231ms | ▄▄▄▄▄▄▄▄ | 2
21.495807ms | ▄▄▄▄ | 1
22.544383ms | | 0
23.592959ms | | 0
24.641535ms | ▄▄▄▄ | 1
25.690111ms | ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ | 4
26.738687ms | ▄▄▄▄ | 1