# HG changeset patch
# User Laman
# Date 2020-04-10 21:45:46
# Node ID 329ff9ed7905961dd2c9e55f3b8d38401904edf1
# Parent  37a1df17b9a1dd0295168b526076d083ca4ea9f0

added benchmark script

diff --git a/readme.md b/readme.md
--- a/readme.md
+++ b/readme.md
@@ -1,7 +1,42 @@
 # Shamira #
 
-Implements Shamir's secret sharing algorithm. Splits a string or a byte sequence byte-per-byte into n<255 shares, with any k of them sufficient for reconstruction of the original input.
+Implements Shamir's secret sharing algorithm. Splits a string or a byte sequence byte-per-byte into _n_<255 shares, with any _k_ of them sufficient for reconstruction of the original input.
 
 Outputs the shares as hexadecimal, Base32 or Base64 encoded strings.
 
-Can be used on its own from the command line by invoking shamira.py or as a library by importing shamira.py.
+Can be used on its own from the command line by invoking `shamira.py` or as a library by importing `shamira.py`.
+
+## Performance ##
+
+As it is, the code is not very fast. Splitting takes _n_ evaluations of a polynomial of order _k_ over Galois field 256, leading to _O(n*k)_ finite field multiplications. Reconstruction of the constant parameters during joining similarly takes _O(k*k)_ multiplications.
+
+Benchmark results, all values mean _seconds per byte_ of the secret length:
+<table>
+    <tr>
+        <th>k / n parameters</th>
+        <th>Split</th>
+        <th>Join</th>
+    </tr>
+    <tr>
+        <td>2 / 3 (a Raspberry Pi 3)</td>
+        <td>7.99e-05</td>
+        <td>0.000428</td>
+    </tr>
+    <tr>
+        <td>2 / 3 (a laptop)</td>
+        <td>1e-05</td>
+        <td>6.7e-05</td>
+    </tr>
+    <tr>
+        <td>254 / 254 (a Raspberry Pi 3)</td>
+        <td>0.417</td>
+        <td>0.471</td>
+    </tr>
+    <tr>
+        <td>254 / 254 (a laptop)</td>
+        <td>0.0431</td>
+        <td>0.0542</td>
+    </tr>
+</table>
+
+While the speeds are not awful, for longer secrets I recommend encrypting them with a random key of your choice and splitting only the key. Anyway, you can run your own benchmark with `benchmark.py`
diff --git a/src/benchmark.py b/src/benchmark.py
new file mode 100644
--- /dev/null
+++ b/src/benchmark.py
@@ -0,0 +1,41 @@
+from argparse import ArgumentParser
+import cProfile
+import timeit
+
+from shamira import generate, reconstruct
+from tests.test_shamira import TestShamira
+
+
+def measure(args):
+	secret = "1234567890123456"
+	shares = generate(secret, args.k, args.n)
+	symbols = globals()
+	symbols.update(locals())
+
+	time = timeit.timeit("""generate(secret, args.k, args.n)""", number=1, globals=symbols)
+	print("The generation took {0:.3}s, {1:.3} per byte.".format(time, time/16))
+
+	time = timeit.timeit("""reconstruct(*shares)""", number=1, globals=symbols)
+	print("The reconstruction took {0:.3}s, {1:.3} per byte.".format(time, time/16))
+
+
+def profile(args):
+	t = TestShamira()
+
+	cProfile.runctx(r"""t.test_generate_reconstruct()""", globals=globals(), locals=locals())
+
+
+parser = ArgumentParser()
+parser.set_defaults(func=lambda _: parser.error("missing command"))
+subparsers = parser.add_subparsers()
+
+profile_parser = subparsers.add_parser("profile")
+profile_parser.set_defaults(func=profile)
+
+measure_parser = subparsers.add_parser("measure")
+measure_parser.add_argument("-k", type=int, required=True)
+measure_parser.add_argument("-n", type=int, required=True)
+measure_parser.set_defaults(func=measure)
+
+args = parser.parse_args()
+args.func(args)