summaryrefslogtreecommitdiff
path: root/scripts/benchmark/legacy-vs-blobs/README.rst
diff options
context:
space:
mode:
authordrebs <drebs@riseup.net>2017-09-13 11:39:36 -0300
committerdrebs <drebs@riseup.net>2017-09-14 12:14:44 -0300
commitdd7b2e414adf2c84873debcd7b7d526859036bc2 (patch)
tree57f44bd454e9f4612a1c4fc808fbb944d0716c18 /scripts/benchmark/legacy-vs-blobs/README.rst
parentaffc09e18a243b215dd1430f53c74a46eab8c098 (diff)
[test] move legacy-vs-blobs graphing to benchmark scripts dir
Diffstat (limited to 'scripts/benchmark/legacy-vs-blobs/README.rst')
-rw-r--r--scripts/benchmark/legacy-vs-blobs/README.rst43
1 files changed, 43 insertions, 0 deletions
diff --git a/scripts/benchmark/legacy-vs-blobs/README.rst b/scripts/benchmark/legacy-vs-blobs/README.rst
new file mode 100644
index 00000000..3eedb3a4
--- /dev/null
+++ b/scripts/benchmark/legacy-vs-blobs/README.rst
@@ -0,0 +1,43 @@
+Ploggin Legacy vs Blobs results
+===============================
+
+This script produces a bar graph comparing different implementations of the
+mail pipeline (legacy, blobs, blobs with session cache, blobs with session
+cache and persistent http connections) for different amount/sizes of incoming
+data (1 x 10M, 10 x 1M, 100 x 100K, 1000 x 10K).
+
+Data acquisition
+----------------
+
+Data was gathered through different runs of the code in
+`soledad/tests/benchmarks/test_legacy_vs_blobs.py`. As these tests are skipped
+by default, you have to (1) remove the @pytest.mark.skip decorator and (2)
+remove the second call to pytest in `tox.ini` that would cause a second run of
+the tests for measuring memory. This script only plots time, for now.
+
+Once you done the above, chdir into `soledad/tests` and do something like this,
+for each test group:
+
+ tox -e benchmarks -- -m benchmark_test_legacy_vs_blobs_10_1000k --benchmark-autosave
+
+Each run of tox as above produces a JSON file in `soledad/tests/.benchmarks`
+with stats. Each file has to be copied in a subdirectory of the `data/`
+directory of the folder where this script lies. Code has to be adapted for each
+extra set of results added to the `data/` directory.
+
+Currently, we are comparing 4 implementation variations:
+
+ - data/no-cache: contains legacy as baseline and blobs as the current master
+ implementation of blobs.
+ - data/cache: adds twisted session cache to blobs connections.
+ - data/persistent: adds session cache and persistent HTTP connections.
+
+Plotting
+--------
+
+Once you have all your data in subdirectories of the `data/` directory and the
+script is correctly tuned for your data, running `make` should be enough to
+create a virtualenvironment with dependencies and plotting data.
+
+By default, the script will generate a file called `./legacy-vs-blobs.png` in
+the current directory with the plot of the data.