Awesome
Fast Elixir
There is a wonderful project in Ruby called fast-ruby, from which I got the inspiration for this repo. The idea is to collect various idioms for writing performant code when there is more than one essentially symantically identical way of computing something. There may be slight differences, so please be sure that when you're changing something that it doesn't change the correctness of your program.
Each idiom has a corresponding code example that resides in code.
Let's write faster code, together! <3
Measurement Tool
We use benchee.
Contributing
Help us collect benchmarks! Please read the contributing guide.
Idioms
- Map Lookup vs. Pattern Matching Lookup
- IO Lists vs. String Concatenation
- Combining lists with
|vs.++ - Putting into maps with
Map.putandput_in - Splitting Strings
sortvs.sort_by- Retrieving state from ets tables vs. Gen Servers
- Writing state in ets tables, persistent_term and Gen Servers
- Comparing strings vs. atoms
- spawn vs. spawn_link
- Replacements for Enum.filter_map/3
- Filtering maps
Map Lookup vs. Pattern Matching Lookup code
If you need to lookup static values in a key-value based structure, you might at first consider assigning a map as a module attribute and looking that up. However, it's significantly faster to use pattern matching to define functions that behave like a key-value based data structure.
$ mix run code/general/map_lookup_vs_pattern_matching.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: none specified
Estimated total run time: 24 s
Benchmarking Map Lookup...
Benchmarking Pattern Matching...
Name ips average deviation median 99th %
Pattern Matching 909.12 K 1.10 μs ±3606.70% 1 μs 2 μs
Map Lookup 792.96 K 1.26 μs ±532.10% 1 μs 2 μs
Comparison:
Pattern Matching 909.12 K
Map Lookup 792.96 K - 1.15x slower +0.161 μs
IO Lists vs. String Concatenation code
Chances are, eventually you'll need to concatenate strings for some sort of output. This could be in a web response, a CLI output, or writing to a file. The faster way to do this is to use IO Lists rather than string concatenation or interpolation.
$ mix run code/general/io_lists_vs_concatenation.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: 100 3-character strings, 100 300-character strings, 5 3-character_strings, 5 300-character_strings, 50 3-character strings, 50 300-character strings
Estimated total run time: 2.40 min
Benchmarking IO List with input 100 3-character strings...
Benchmarking IO List with input 100 300-character strings...
Benchmarking IO List with input 5 3-character_strings...
Benchmarking IO List with input 5 300-character_strings...
Benchmarking IO List with input 50 3-character strings...
Benchmarking IO List with input 50 300-character strings...
Benchmarking Interpolation with input 100 3-character strings...
Benchmarking Interpolation with input 100 300-character strings...
Benchmarking Interpolation with input 5 3-character_strings...
Benchmarking Interpolation with input 5 300-character_strings...
Benchmarking Interpolation with input 50 3-character strings...
Benchmarking Interpolation with input 50 300-character strings...
##### With input 100 3-character strings #####
Name ips average deviation median 99th %
IO List 1.41 M 0.71 μs ±4475.40% 1 μs 2 μs
Interpolation 0.31 M 3.27 μs ±76.91% 3 μs 11 μs
Comparison:
IO List 1.41 M
Interpolation 0.31 M - 4.61x slower +2.56 μs
##### With input 100 300-character strings #####
Name ips average deviation median 99th %
IO List 1.40 M 0.71 μs ±4411.36% 1 μs 1 μs
Interpolation 0.20 M 4.90 μs ±248.22% 4 μs 22 μs
Comparison:
IO List 1.40 M
Interpolation 0.20 M - 6.86x slower +4.18 μs
##### With input 5 3-character_strings #####
Name ips average deviation median 99th %
IO List 5.15 M 194.15 ns ±2555.27% 0 ns 1000 ns
Interpolation 1.84 M 544.12 ns ±4764.73% 0 ns 2000 ns
Comparison:
IO List 5.15 M
Interpolation 1.84 M - 2.80x slower +349.96 ns
##### With input 5 300-character_strings #####
Name ips average deviation median 99th %
IO List 5.03 M 198.76 ns ±4663.45% 0 ns 1000 ns
Interpolation 1.92 M 521.81 ns ±193.09% 0 ns 1000 ns
Comparison:
IO List 5.03 M
Interpolation 1.92 M - 2.63x slower +323.05 ns
##### With input 50 3-character strings #####
Name ips average deviation median 99th %
IO List 1.94 M 0.52 μs ±6397.19% 0 μs 2 μs
Interpolation 0.57 M 1.75 μs ±130.98% 2 μs 2 μs
Comparison:
IO List 1.94 M
Interpolation 0.57 M - 3.40x slower +1.24 μs
##### With input 50 300-character strings #####
Name ips average deviation median 99th %
IO List 2.06 M 0.49 μs ±8825.39% 0 μs 2 μs
Interpolation 0.37 M 2.71 μs ±657.41% 2 μs 14 μs
Comparison:
IO List 2.06 M
Interpolation 0.37 M - 5.58x slower +2.22 μs
Combining lists with | vs. ++ code
Adding two lists together might seem like a simple problem to solve, but in
Elixir there are a couple ways to solve that issue. We can use ++ to
concatenate two lists easily: [1, 2] ++ [3, 4] #=> [1, 2, 3, 4], but the
problem with that approach is that once you start dealing with larger lists it
becomes VERY slow! Because of this, when combining two lists, you should try
and use the cons operator (|) whenever possible. This will require you to
remember to flatten the resulting nested list, but it's a huge performance
optimization on larger lists.
$ mix run ./code/general/concat_vs_cons.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: 1,000 large items, 1,000 small items, 10 large items, 10 small items, 100 large items, 100 small items
Estimated total run time: 3.60 min
Benchmarking Concatenation with input 1,000 large items...
Benchmarking Concatenation with input 1,000 small items...
Benchmarking Concatenation with input 10 large items...
Benchmarking Concatenation with input 10 small items...
Benchmarking Concatenation with input 100 large items...
Benchmarking Concatenation with input 100 small items...
Benchmarking Cons + Flatten with input 1,000 large items...
Benchmarking Cons + Flatten with input 1,000 small items...
Benchmarking Cons + Flatten with input 10 large items...
Benchmarking Cons + Flatten with input 10 small items...
Benchmarking Cons + Flatten with input 100 large items...
Benchmarking Cons + Flatten with input 100 small items...
Benchmarking Cons + Reverse + Flatten with input 1,000 large items...
Benchmarking Cons + Reverse + Flatten with input 1,000 small items...
Benchmarking Cons + Reverse + Flatten with input 10 large items...
Benchmarking Cons + Reverse + Flatten with input 10 small items...
Benchmarking Cons + Reverse + Flatten with input 100 large items...
Benchmarking Cons + Reverse + Flatten with input 100 small items...
##### With input 1,000 large items #####
Name ips average deviation median 99th %
Cons + Reverse + Flatten 38.45 26.01 ms ±6.11% 25.91 ms 30.56 ms
Cons + Flatten 38.38 26.06 ms ±6.39% 26.06 ms 29.32 ms
Concatenation 0.179 5573.57 ms ±0.26% 5573.57 ms 5583.94 ms
Comparison:
Cons + Reverse + Flatten 38.45
Cons + Flatten 38.38 - 1.00x slower +0.0501 ms
Concatenation 0.179 - 214.32x slower +5547.56 ms
##### With input 1,000 small items #####
Name ips average deviation median 99th %
Cons + Reverse + Flatten 3.78 K 264.27 μs ±19.49% 243 μs 496 μs
Cons + Flatten 3.76 K 266.16 μs ±18.53% 246 μs 491.83 μs
Concatenation 0.0626 K 15984.51 μs ±8.58% 15927 μs 20412.82 μs
Comparison:
Cons + Reverse + Flatten 3.78 K
Cons + Flatten 3.76 K - 1.01x slower +1.90 μs
Concatenation 0.0626 K - 60.49x slower +15720.24 μs
##### With input 10 large items #####
Name ips average deviation median 99th %
Concatenation 8.33 K 120.04 μs ±31.79% 111 μs 268 μs
Cons + Flatten 5.12 K 195.17 μs ±20.09% 181 μs 378 μs
Cons + Reverse + Flatten 5.11 K 195.88 μs ±20.32% 181 μs 378 μs
Comparison:
Concatenation 8.33 K
Cons + Flatten 5.12 K - 1.63x slower +75.13 μs
Cons + Reverse + Flatten 5.11 K - 1.63x slower +75.85 μs
##### With input 10 small items #####
Name ips average deviation median 99th %
Concatenation 575.41 K 1.74 μs ±1951.31% 1 μs 4 μs
Cons + Flatten 331.62 K 3.02 μs ±972.07% 3 μs 7 μs
Cons + Reverse + Flatten 330.05 K 3.03 μs ±853.79% 3 μs 8 μs
Comparison:
Concatenation 575.41 K
Cons + Flatten 331.62 K - 1.74x slower +1.28 μs
Cons + Reverse + Flatten 330.05 K - 1.74x slower +1.29 μs
##### With input 100 large items #####
Name ips average deviation median 99th %
Cons + Reverse + Flatten 38.56 25.93 ms ±6.25% 25.85 ms 32.02 ms
Cons + Flatten 38.35 26.08 ms ±6.30% 26.04 ms 30.68 ms
Concatenation 0.180 5561.40 ms ±0.41% 5561.40 ms 5577.71 ms
Comparison:
Cons + Reverse + Flatten 38.56
Cons + Flatten 38.35 - 1.01x slower +0.145 ms
Concatenation 0.180 - 214.47x slower +5535.47 ms
##### With input 100 small items #####
Name ips average deviation median 99th %
Cons + Flatten 38.68 K 25.85 μs ±32.87% 24 μs 69 μs
Cons + Reverse + Flatten 38.23 K 26.16 μs ±39.65% 24 μs 70 μs
Concatenation 4.33 K 230.99 μs ±50.47% 213 μs 590.06 μs
Comparison:
Cons + Flatten 38.68 K
Cons + Reverse + Flatten 38.23 K - 1.01x slower +0.31 μs
Concatenation 4.33 K - 8.94x slower +205.13 μs
Putting into maps with Map.put and put_in code
Do not put data into root of map with put_in. It is ~2x slower than Map.put. Also put_in/2
is more effective than put_in/3.
$ mix run ./code/general/map_put_vs_put_in.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: Large (30,000 items), Medium (3,000 items), Small (30 items)
Estimated total run time: 1.80 min
Benchmarking Map.put/3 with input Large (30,000 items)...
Benchmarking Map.put/3 with input Medium (3,000 items)...
Benchmarking Map.put/3 with input Small (30 items)...
Benchmarking put_in/2 with input Large (30,000 items)...
Benchmarking put_in/2 with input Medium (3,000 items)...
Benchmarking put_in/2 with input Small (30 items)...
Benchmarking put_in/3 with input Large (30,000 items)...
Benchmarking put_in/3 with input Medium (3,000 items)...
Benchmarking put_in/3 with input Small (30 items)...
##### With input Large (30,000 items) #####
Name ips average deviation median 99th %
Map.put/3 247.43 4.04 ms ±10.45% 3.97 ms 5.41 ms
put_in/2 242.10 4.13 ms ±12.48% 4.01 ms 5.74 ms
put_in/3 221.53 4.51 ms ±11.11% 4.41 ms 6.13 ms
Comparison:
Map.put/3 247.43
put_in/2 242.10 - 1.02x slower +0.0888 ms
put_in/3 221.53 - 1.12x slower +0.47 ms
##### With input Medium (3,000 items) #####
Name ips average deviation median 99th %
Map.put/3 5.68 K 175.98 μs ±34.49% 150.98 μs 400.98 μs
put_in/2 3.62 K 276.42 μs ±23.76% 252.98 μs 546.98 μs
put_in/3 3.09 K 323.22 μs ±22.44% 296.98 μs 630.98 μs
Comparison:
Map.put/3 5.68 K
put_in/2 3.62 K - 1.57x slower +100.44 μs
put_in/3 3.09 K - 1.84x slower +147.23 μs
##### With input Small (30 items) #####
Name ips average deviation median 99th %
Map.put/3 1040.86 K 0.96 μs ±3795.74% 0.98 μs 1.98 μs
put_in/2 400.53 K 2.50 μs ±1295.21% 1.98 μs 2.98 μs
put_in/3 338.63 K 2.95 μs ±1124.35% 1.98 μs 3.98 μs
Comparison:
Map.put/3 1040.86 K
put_in/2 400.53 K - 2.60x slower +1.54 μs
put_in/3 338.63 K - 3.07x slower +1.99 μs
Splitting Large Strings code
Elixir's String.split/2 is the fastest option for splitting strings by far, but
using a String literal as the splitter instead of a regex will yield significant
performance benefits.
$ mix run code/general/string_split_large_strings.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: Large string (1 Million Numbers), Medium string (10 Thousand Numbers), Small string (1 Hundred Numbers)
Estimated total run time: 2.40 min
Benchmarking split with input Large string (1 Million Numbers)...
Benchmarking split with input Medium string (10 Thousand Numbers)...
Benchmarking split with input Small string (1 Hundred Numbers)...
Benchmarking split erlang with input Large string (1 Million Numbers)...
Benchmarking split erlang with input Medium string (10 Thousand Numbers)...
Benchmarking split erlang with input Small string (1 Hundred Numbers)...
Benchmarking split regex with input Large string (1 Million Numbers)...
Benchmarking split regex with input Medium string (10 Thousand Numbers)...
Benchmarking split regex with input Small string (1 Hundred Numbers)...
Benchmarking splitter |> to_list with input Large string (1 Million Numbers)...
Benchmarking splitter |> to_list with input Medium string (10 Thousand Numbers)...
Benchmarking splitter |> to_list with input Small string (1 Hundred Numbers)...
##### With input Large string (1 Million Numbers) #####
Name ips average deviation median 99th %
split 13.96 71.63 ms ±29.57% 59.81 ms 121.28 ms
splitter |> to_list 3.24 308.26 ms ±14.54% 290.97 ms 442.09 ms
split erlang 1.09 919.28 ms ±4.86% 939.75 ms 998.24 ms
split regex 0.78 1286.40 ms ±9.80% 1253.48 ms 1489.63 ms
Comparison:
split 13.96
splitter |> to_list 3.24 - 4.30x slower +236.62 ms
split erlang 1.09 - 12.83x slower +847.65 ms
split regex 0.78 - 17.96x slower +1214.77 ms
##### With input Medium string (10 Thousand Numbers) #####
Name ips average deviation median 99th %
split 3813.15 0.26 ms ±45.13% 0.21 ms 0.57 ms
splitter |> to_list 397.04 2.52 ms ±14.65% 2.48 ms 3.73 ms
split erlang 137.55 7.27 ms ±8.52% 7.17 ms 9.35 ms
split regex 93.73 10.67 ms ±7.46% 10.56 ms 13.07 ms
Comparison:
split 3813.15
splitter |> to_list 397.04 - 9.60x slower +2.26 ms
split erlang 137.55 - 27.72x slower +7.01 ms
split regex 93.73 - 40.68x slower +10.41 ms
##### With input Small string (1 Hundred Numbers) #####
Name ips average deviation median 99th %
split 365.94 K 2.73 μs ±634.81% 2 μs 14 μs
splitter |> to_list 45.63 K 21.92 μs ±45.25% 20 μs 63 μs
split erlang 14.19 K 70.48 μs ±48.03% 53 μs 186.91 μs
split regex 9.87 K 101.28 μs ±24.68% 93 μs 222 μs
Comparison:
split 365.94 K
splitter |> to_list 45.63 K - 8.02x slower +19.18 μs
split erlang 14.19 K - 25.79x slower +67.74 μs
split regex 9.87 K - 37.06x slower +98.55 μs
sort vs. sort_by code
Sorting a list of maps or keyword lists can be done in various ways. However, since the sort
behavior is fairly implicit if you're sorting without a defined sort function, and since the
speed difference is quite small, it's probably best to use sort/2 or sort_by/2 in all
cases when sorting lists and maps (including keyword lists and structs).
$ mix run code/general/sort_vs_sort_by.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: none specified
Estimated total run time: 36 s
Benchmarking sort/1...
Benchmarking sort/2...
Benchmarking sort_by/2...
Name ips average deviation median 99th %
sort/1 7.82 K 127.86 μs ±23.45% 118 μs 269 μs
sort/2 7.01 K 142.57 μs ±22.48% 132 μs 294 μs
sort_by/2 6.68 K 149.62 μs ±22.70% 138 μs 308 μs
Comparison:
sort/1 7.82 K
sort/2 7.01 K - 1.12x slower +14.71 μs
sort_by/2 6.68 K - 1.17x slower +21.76 μs
Retrieving state from ets tables vs. Gen Servers code
There are many differences between Gen Servers and ets tables, but many people have often praised ets tables for being extremely fast. For the simple case of retrieving information from a key-value store, the ets table is indeed much faster for reads. For more complicated use cases, and for comparisons of writes instead of reads, further benchmarks are needed, but so far ets lives up to its reputation for speed.
$ mix run code/general/ets_vs_gen_server.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: none specified
Estimated total run time: 24 s
Benchmarking ets table...
Benchmarking gen server...
Name ips average deviation median 99th %
ets table 5.11 M 0.196 μs ±8972.86% 0 μs 0.98 μs
gen server 0.55 M 1.82 μs ±997.04% 1.98 μs 2.98 μs
Comparison:
ets table 5.11 M
gen server 0.55 M - 9.31x slower +1.63 μs
Writing state in ets tables, persistent_term and Gen Servers code
Not only is it faster to read from ets or persistent_term versus a GenServer, but it's also
much faster to write state in these two options. If you have need for state that needs to be
stored but without a lot of behavior around that state, ets or persistent_term is always going
to be the better choice over a GenServer. persistent_term is the fastest to read from by far,
but is global across the VM and also slower to write to, so in most cases ets will be the best
choice for storing state and should be the default option to start with.
$ mix run code/general/ets_vs_gen_server_write.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: none specified
Estimated total run time: 36 s
Benchmarking ets table...
Benchmarking gen server...
Benchmarking persistent term...
Name ips average deviation median 99th %
ets table 5.22 M 191.61 ns ±798.69% 0 ns 1000 ns
persistent term 2.43 M 410.87 ns ±11324.51% 0 ns 1000 ns
gen server 0.58 M 1715.61 ns ±367.31% 2000 ns 2000 ns
Comparison:
ets table 5.22 M
persistent term 2.43 M - 2.14x slower +219.26 ns
gen server 0.58 M - 8.95x slower +1524.00 ns
Comparing strings vs. atoms code
Because atoms are stored in a special table in the BEAM, comparing atoms is rather fast compared to comparing strings, where you need to compare each part of the list that underlies the string. When you have a choice of what type to use, atoms is the faster choice. However, what you probably should not do is to convert strings to atoms solely for the perceived speed benefit, since it ends up being much slower than just comparing the strings, even dozens of times.
$ mix run code/general/comparing_strings_vs_atoms.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 0 ns
parallel: 1
inputs: Large (1-100), Medium (1-50), Small (1-5)
Estimated total run time: 1.80 min
Benchmarking Comparing atoms with input Large (1-100)...
Benchmarking Comparing atoms with input Medium (1-50)...
Benchmarking Comparing atoms with input Small (1-5)...
Benchmarking Comparing strings with input Large (1-100)...
Benchmarking Comparing strings with input Medium (1-50)...
Benchmarking Comparing strings with input Small (1-5)...
Benchmarking Converting to atoms and then comparing with input Large (1-100)...
Benchmarking Converting to atoms and then comparing with input Medium (1-50)...
Benchmarking Converting to atoms and then comparing with input Small (1-5)...
##### With input Large (1-100) #####
Name ips average deviation median 99th %
Comparing atoms 3.74 M 267.46 ns ±12198.11% 0 ns 1000 ns
Comparing strings 3.71 M 269.25 ns ±11719.28% 0 ns 1000 ns
Converting to atoms and then comparing 0.94 M 1065.67 ns ±290.55% 1000 ns 2000 ns
Comparison:
Comparing atoms 3.74 M
Comparing strings 3.71 M - 1.01x slower +1.79 ns
Converting to atoms and then comparing 0.94 M - 3.98x slower +798.21 ns
##### With input Medium (1-50) #####
Name ips average deviation median 99th %
Comparing atoms 3.70 M 270.08 ns ±11419.92% 0 ns 1000 ns
Comparing strings 3.68 M 271.52 ns ±11603.67% 0 ns 1000 ns
Converting to atoms and then comparing 1.34 M 743.76 ns ±2924.56% 1000 ns 1000 ns
Comparison:
Comparing atoms 3.70 M
Comparing strings 3.68 M - 1.01x slower +1.44 ns
Converting to atoms and then comparing 1.34 M - 2.75x slower +473.68 ns
##### With input Small (1-5) #####
Name ips average deviation median 99th %
Comparing atoms 3.81 M 262.27 ns ±11438.39% 0 ns 1000 ns
Comparing strings 3.69 M 270.86 ns ±11945.32% 0 ns 1000 ns
Converting to atoms and then comparing 2.45 M 407.62 ns ±8371.44% 0 ns 1000 ns
Comparison:
Comparing atoms 3.81 M
Comparing strings 3.69 M - 1.03x slower +8.59 ns
Converting to atoms and then comparing 2.45 M - 1.55x slower +145.34 ns
spawn vs. spawn_link code
There are two ways to spawn a process on the BEAM, spawn and spawn_link.
Because spawn_link links the child process to the process which spawned it, it
takes slightly longer. The way in which processes are spawned is unlikely to be
a bottleneck in most applications, though, and the resiliency benefits of OTP
supervision trees vastly outweighs the slightly slower run time of spawn_link,
so that should still be favored in nearly every case in which processes need to
be spawned.
$ mix run code/general/spawn_vs_spawn_link.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 2 s
parallel: 1
inputs: none specified
Estimated total run time: 28 s
Benchmarking spawn/1...
Benchmarking spawn_link/1...
Name ips average deviation median 99th %
spawn/1 636.00 K 1.57 μs ±1512.39% 1 μs 2 μs
spawn_link/1 576.18 K 1.74 μs ±1402.58% 2 μs 2 μs
Comparison:
spawn/1 636.00 K
spawn_link/1 576.18 K - 1.10x slower +0.163 μs
Memory usage statistics:
Name Memory usage
spawn/1 72 B
spawn_link/1 72 B - 1.00x memory usage +0 B
**All measurements for memory usage were the same**
Replacements for Enum.filter_map/3 code
Elixir used to have an Enum.filter_map/3 function that would filter a list and
also apply a function to each element in the list that was not removed, but it
was deprecated in version 1.5. Luckily there are still four other ways to do
that same thing! They're all mostly the same, but if you're looking for the
options with the best performance your best bet is to use either a for
comprehension or Enum.reduce/3 and then Enum.reverse/1. Using
Enum.filter/2 and then Enum.map/2 is also a fine choice, but it has higher
memory usage than the other two options.
The one option you should avoid is using Enum.flat_map/2 as it is both slower
and has higher memory usage.
$ mix run code/general/filter_map.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 10 ms
parallel: 1
inputs: Large, Medium, Small
Estimated total run time: 2.40 min
Benchmarking filter |> map with input Large...
Benchmarking filter |> map with input Medium...
Benchmarking filter |> map with input Small...
Benchmarking flat_map with input Large...
Benchmarking flat_map with input Medium...
Benchmarking flat_map with input Small...
Benchmarking for comprehension with input Large...
Benchmarking for comprehension with input Medium...
Benchmarking for comprehension with input Small...
Benchmarking reduce |> reverse with input Large...
Benchmarking reduce |> reverse with input Medium...
Benchmarking reduce |> reverse with input Small...
##### With input Large #####
Name ips average deviation median 99th %
reduce |> reverse 12.12 82.51 ms ±4.60% 81.46 ms 97.24 ms
for comprehension 12.12 82.51 ms ±4.53% 81.87 ms 94.38 ms
filter |> map 10.78 92.75 ms ±4.91% 92.15 ms 103.58 ms
flat_map 8.41 118.89 ms ±3.22% 118.22 ms 134.28 ms
Comparison:
reduce |> reverse 12.12
for comprehension 12.12 - 1.00x slower +0.00348 ms
filter |> map 10.78 - 1.12x slower +10.24 ms
flat_map 8.41 - 1.44x slower +36.38 ms
Memory usage statistics:
Name Memory usage
reduce |> reverse 7.57 MB
for comprehension 7.57 MB - 1.00x memory usage +0 MB
filter |> map 13.28 MB - 1.75x memory usage +5.71 MB
flat_map 14.32 MB - 1.89x memory usage +6.75 MB
**All measurements for memory usage were the same**
##### With input Medium #####
Name ips average deviation median 99th %
for comprehension 1.27 K 788.69 μs ±14.54% 732 μs 1287.38 μs
reduce |> reverse 1.26 K 792.37 μs ±14.73% 732 μs 1283.97 μs
filter |> map 1.16 K 859.07 μs ±14.68% 802 μs 1377.75 μs
flat_map 0.86 K 1157.55 μs ±15.68% 1093 μs 1838.80 μs
Comparison:
for comprehension 1.27 K
reduce |> reverse 1.26 K - 1.00x slower +3.68 μs
filter |> map 1.16 K - 1.09x slower +70.38 μs
flat_map 0.86 K - 1.47x slower +368.87 μs
Memory usage statistics:
Name Memory usage
for comprehension 57.13 KB
reduce |> reverse 57.13 KB - 1.00x memory usage +0 KB
filter |> map 109.12 KB - 1.91x memory usage +51.99 KB
flat_map 130.66 KB - 2.29x memory usage +73.54 KB
**All measurements for memory usage were the same**
##### With input Small #####
Name ips average deviation median 99th %
reduce |> reverse 121.39 K 8.24 μs ±179.26% 8 μs 30 μs
for comprehension 121.20 K 8.25 μs ±180.01% 8 μs 30 μs
filter |> map 111.29 K 8.99 μs ±144.77% 8 μs 31 μs
flat_map 85.08 K 11.75 μs ±119.95% 11 μs 37 μs
Comparison:
reduce |> reverse 121.39 K
for comprehension 121.20 K - 1.00x slower +0.0133 μs
filter |> map 111.29 K - 1.09x slower +0.75 μs
flat_map 85.08 K - 1.43x slower +3.52 μs
Memory usage statistics:
Name Memory usage
reduce |> reverse 1.09 KB
for comprehension 1.09 KB - 1.00x memory usage +0 KB
filter |> map 1.60 KB - 1.46x memory usage +0.51 KB
flat_map 1.62 KB - 1.48x memory usage +0.52 KB
**All measurements for memory usage were the same**
String.slice/3 vs :binary.part/3 code
From String.slice/3 documentation:
Remember this function works with Unicode graphemes and considers the slices to represent grapheme offsets. If you want to split on raw bytes, check Kernel.binary_part/3 instead.
$ mix run code/general/string_slice.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 100 ms
time: 2 s
memory time: 10 ms
parallel: 1
inputs: Large string (10 Thousand Numbers), Small string (10 Numbers)
Estimated total run time: 12.66 s
Benchmarking :binary.part/3 with input Large string (10 Thousand Numbers)...
Benchmarking :binary.part/3 with input Small string (10 Numbers)...
Benchmarking String.slice/3 with input Large string (10 Thousand Numbers)...
Benchmarking String.slice/3 with input Small string (10 Numbers)...
Benchmarking binary_part/3 with input Large string (10 Thousand Numbers)...
Benchmarking binary_part/3 with input Small string (10 Numbers)...
##### With input Large string (10 Thousand Numbers) #####
Name ips average deviation median 99th %
binary_part/3 11.14 M 89.78 ns ±2513.45% 100 ns 200 ns
:binary.part/3 3.59 M 278.65 ns ±9466.55% 0 ns 1000 ns
String.slice/3 0.90 M 1112.12 ns ±440.40% 1000 ns 2000 ns
Comparison:
binary_part/3 11.14 M
:binary.part/3 3.59 M - 3.10x slower +188.87 ns
String.slice/3 0.90 M - 12.39x slower +1022.34 ns
Memory usage statistics:
Name Memory usage
binary_part/3 0 B
:binary.part/3 0 B - 1.00x memory usage +0 B
String.slice/3 880 B - ∞ x memory usage +880 B
**All measurements for memory usage were the same**
##### With input Small string (10 Numbers) #####
Name ips average deviation median 99th %
binary_part/3 3.64 M 274.57 ns ±7776.31% 0 ns 1000 ns
:binary.part/3 3.56 M 281.06 ns ±9071.16% 0 ns 1000 ns
String.slice/3 0.91 M 1103.31 ns ±246.39% 1000 ns 2000 ns
Comparison:
binary_part/3 3.64 M
:binary.part/3 3.56 M - 1.02x slower +6.48 ns
String.slice/3 0.91 M - 4.02x slower +828.73 ns
Memory usage statistics:
Name Memory usage
binary_part/3 0 B
:binary.part/3 0 B - 1.00x memory usage +0 B
String.slice/3 880 B - ∞ x memory usage +880 B
**All measurements for memory usage were the same**
Filtering maps code
If we have a map and want to filter out key-value pairs from that map, there are
several ways to do it. However, because of some optimizations in Erlang,
:maps.filter/2 is faster than any of the versions implemented in Elixir.
If you look at the benchmark code, you'll notice that the function used for
filtering takes two arguments (the key and value) instead of one (a tuple with
the key and value), and it's this difference that is responsible for the
decreased execution time and memory usage.
$ mix run code/general/filtering_maps.exs
Operating System: macOS
CPU Information: Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz
Number of Available Cores: 16
Available memory: 16 GB
Elixir 1.11.0-rc.0
Erlang 23.0.2
Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 1 s
parallel: 1
inputs: Large (10_000), Medium (100), Small (1)
Estimated total run time: 2.60 min
Benchmarking :maps.filter with input Large (10_000)...
Benchmarking :maps.filter with input Medium (100)...
Benchmarking :maps.filter with input Small (1)...
Benchmarking Enum.filter/2 |> Enum.into/2 with input Large (10_000)...
Benchmarking Enum.filter/2 |> Enum.into/2 with input Medium (100)...
Benchmarking Enum.filter/2 |> Enum.into/2 with input Small (1)...
Benchmarking Enum.filter/2 |> Map.new/1 with input Large (10_000)...
Benchmarking Enum.filter/2 |> Map.new/1 with input Medium (100)...
Benchmarking Enum.filter/2 |> Map.new/1 with input Small (1)...
Benchmarking for with input Large (10_000)...
Benchmarking for with input Medium (100)...
Benchmarking for with input Small (1)...
##### With input Large (10_000) #####
Name ips average deviation median 99th %
:maps.filter 669.86 1.49 ms ±14.38% 1.44 ms 2.31 ms
Enum.filter/2 |> Enum.into/2 532.59 1.88 ms ±19.86% 1.78 ms 2.87 ms
Enum.filter/2 |> Map.new/1 527.37 1.90 ms ±25.17% 1.79 ms 2.85 ms
for 524.51 1.91 ms ±31.33% 1.80 ms 2.83 ms
Comparison:
:maps.filter 669.86
Enum.filter/2 |> Enum.into/2 532.59 - 1.26x slower +0.38 ms
Enum.filter/2 |> Map.new/1 527.37 - 1.27x slower +0.40 ms
for 524.51 - 1.28x slower +0.41 ms
Memory usage statistics:
Name Memory usage
:maps.filter 780.45 KB
Enum.filter/2 |> Enum.into/2 782.85 KB - 1.00x memory usage +2.41 KB
Enum.filter/2 |> Map.new/1 782.87 KB - 1.00x memory usage +2.42 KB
for 782.86 KB - 1.00x memory usage +2.41 KB
**All measurements for memory usage were the same**
##### With input Medium (100) #####
Name ips average deviation median 99th %
:maps.filter 76.01 K 13.16 μs ±90.13% 12 μs 42 μs
Enum.filter/2 |> Map.new/1 61.19 K 16.34 μs ±61.27% 15 μs 50 μs
for 60.89 K 16.42 μs ±65.36% 15 μs 51 μs
Enum.filter/2 |> Enum.into/2 60.60 K 16.50 μs ±60.52% 15 μs 51 μs
Comparison:
:maps.filter 76.01 K
Enum.filter/2 |> Map.new/1 61.19 K - 1.24x slower +3.19 μs
for 60.89 K - 1.25x slower +3.27 μs
Enum.filter/2 |> Enum.into/2 60.60 K - 1.25x slower +3.35 μs
Memory usage statistics:
Name Memory usage
:maps.filter 5.67 KB
Enum.filter/2 |> Map.new/1 7.84 KB - 1.38x memory usage +2.17 KB
for 7.84 KB - 1.38x memory usage +2.17 KB
Enum.filter/2 |> Enum.into/2 7.84 KB - 1.38x memory usage +2.17 KB
**All measurements for memory usage were the same**
##### With input Small (1) #####
Name ips average deviation median 99th %
:maps.filter 2.46 M 406.55 ns ±6862.02% 0 ns 1000 ns
for 1.81 M 551.70 ns ±4974.10% 0 ns 1000 ns
Enum.filter/2 |> Map.new/1 1.78 M 562.13 ns ±5004.53% 0 ns 1000 ns
Enum.filter/2 |> Enum.into/2 1.64 M 608.18 ns ±4796.51% 1000 ns 1000 ns
Comparison:
:maps.filter 2.46 M
for 1.81 M - 1.36x slower +145.15 ns
Enum.filter/2 |> Map.new/1 1.78 M - 1.38x slower +155.58 ns
Enum.filter/2 |> Enum.into/2 1.64 M - 1.50x slower +201.63 ns
Memory usage statistics:
Name Memory usage
:maps.filter 136 B
for 248 B - 1.82x memory usage +112 B
Enum.filter/2 |> Map.new/1 248 B - 1.82x memory usage +112 B
Enum.filter/2 |> Enum.into/2 248 B - 1.82x memory usage +112 B
**All measurements for memory usage were the same**
Something went wrong
Something look wrong to you? :cry: Have a better example? :heart_eyes: Excellent!
Please open an Issue or open a Pull Request to fix it.
Thank you in advance! :wink: :beer:
Also Checkout
-
Talk by @PragTob from ElixirLive 2016 about benchmarking in Elixir.
-
Wonderful static analysis tool by @rrrene. It's not just about speed, but it will flag some performance issues.
Brought to you by @devoncestes
License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Code License
CC0 1.0 Universal
To the extent possible under law, @devonestes has waived all copyright and related or neighboring rights to "fast-elixir".
This work belongs to the community.