add some benchmarks to README

README.md

@@ -1,7 +1,8 @@
+[[_TOC_]]
 
 # plag
 
-Tools to optimize a list of IP (v4 or v6) prefixes.
+Plag is a couple of tools to optimize a list of IP (v4 or v6) prefixes.
 The tools read prefix list on stdin and output the result to stdout.
 
  - plagmax - returns a minimal superset of prefixes (loose mode)
@@ -57,7 +58,7 @@ Output:
 10.0.2.1/32
 ```
 
-## Building
+# Building
 
 You will need only a C compiler and a standard library to build the tools.
 A simple `make` run should do the job.
@@ -69,7 +70,7 @@ gcc -O3 -o plagmax plagmax.c lib.o
 gcc -O3 -o plageq plageq.c lib.o
 ```
 
-## Running
+# Running
 
 The tools read stdin where they expect IP prefixes one per line.
 Prefixes are of the standard form: `<IP>[/<mask>]`.
@@ -86,8 +87,8 @@ This matches all prefixes, having masks in range `<low>`-`<high>`
 In case `<mask> = <low> = <high>` the submasks part is omitted.
 This notation of submasks is used in `bird` routing daemon.
 
-You can use `bgpq4` tool for example to fetch the list of prefixes
-and pipe its output to the tools:
+You can use [bgpq4](https://github.com/bgp/bgpq4) tool for example
+to fetch the list of prefixes and pipe its output to the tools:
 
 ```bash
 bgpq4 -4 -F '%n/%l\n' as-set-name | ./plageq > as-set-name.ipv4.strict.txt
@@ -95,3 +96,54 @@ bgpq4 -6 -F '%n/%l\n' as-set-name | ./plagmax > as-set-name.ipv6.loose.txt
 ```
 
 Tools also output to stderr some statistics of using preallocated memory.
+
+# Benchmarks
+
+Here are some performance tests. You can see how effective and fast
+the tools are. The times are not throughly measured, so take it as indicative
+only. The tools have also been compared with the analogous functions of
+[bgpq4](https://github.com/bgp/bgpq4) tool.
+It also aggregates and does it amost as fast, and almost as good.
+But still its aggregation is not perfect.
+
+Test sets:
+
+ - ripe{4,6}: route attributes from route objects in the RIPE DB (unsorted)
+ - radb{4,6}: route attributes from route objects in the RADB (unsorted)
+ - ntt{4,6}: AS2914:AS-GLOBAL snapshot taken with bgpq4 (sorted)
+
+Test cases (for bgpq4 a local cache of the real response is used):
+
+ - uniq: `sort -u <source | wc -l`
+ - plageq/plagmax: `plageq <source | wc -l`
+ - bgpq4 (eq): `bgpq4 -4 -A -F '%n/%l{%a,%A}\n' AS-TEST | wc -l` for IPv4
+ - bgpq4 (max): `bgpq4 -4 -R 32 -r 1 -A -F '%n/%l\n' AS-TEST | wc -l` for IPv4
+
+## Number of prefixes
+
+|             |  ripe4  | ripe6  |   radb4   |  radb6  |   ntt4    |  ntt6   |
+| :---------- | ------: | -----: | --------: | ------: | --------: | ------: |
+| source      | 424 813 | 52 506 | 1 351 748 | 166 172 | 2 038 959 | 540 047 |
+| uniq        | 394 241 | 49 568 | 1 086 347 | 149 782 | 2 038 959 | 540 047 |
+| plageq      | 129 678 | 28 657 |   259 805 |  37 919 |   395 831 |  79 875 |
+| bgpq4 (eq)  | 135 690 | 28 830 |   269 235 |  38 341 |   414 558 |  80 764 |
+| plagmax     |  41 447 | 18 636 |    87 643 |  19 733 |    99 794 |  38 023 |
+| bgpq4 (max) |  66 040 | 18 692 |   101 174 |  19 786 |   128 522 |  38 193 |
+
+## Processing time (user)
+
+|             | ripe4 | ripe6 | radb4 | radb6 | ntt4 | ntt6 |
+| :---------- | :---- | :---- | :---- | :---- | :--- | :--- |
+| uniq        |  0.15 |  0.01 |  0.57 |  0.03 | 0.59 | 0.12 |
+| plageq      |  0.19 |  0.05 |  0.53 |  0.12 | 0.87 | 0.31 |
+| bgpq4 (eq)  |  0.36 |  0.05 |  0.99 |  0.15 | 0.94 | 0.3  |
+| plagmax     |  0.1  |  0.02 |  0.31 |  0.03 | 0.37 | 0.15 |
+| bgpq4 (max) |  0.45 |  0.06 |  1.12 |  0.18 | 0.81 | 0.29 |
+
+## Memory usage
+
+|             | ripe4 | ripe6 | radb4 | radb6 | ntt4 | ntt6 |
+| :---------- | ----: | ----: | ----: | ----: | ---: | ---: |
+| source size |  6.4M |  0.9M |   21M |  3.1M |  31M |  10M |
+| plageq      |  8.5M |  2.7M |   21M |  4.8M |  35M |  14M |
+| plagmax     |  1.1M |  1.0M |  2.4M |  1.2M | 2.3M | 2.1M |