In this example, the size of payload is only 13 bytes. Using
JSON.stringify instead causes the payload to be 100 bytes.
If you can't emit message strings and can only send array buffers by themselves, add something like
__message: "uint8" to the start of all your schemas/objects. On the receiver you can just read the first byte of the buffer to determine what message it is.
I was working on an app that used WebSockets to talk between client and server. Usually when doing this the client and server just send JSON back and forth. However, when receiving a message the receiver already knows what the format of the message is going to be. Example:
// Client:var message = 'sender': 'John' 'contents': 'hi' ;socket;// Serversocket;
The problems I had with sending JSON back and forth between client and server:
- It's a complete waste of bandwidth to send all those keys and delimiters when the object format is known.
- Even though
JSON.parseare optimized native functions, they're slower than buffers.
- There's no implicit central message repository where I can look at the format of all my different packets.
- There's no validation so there's potential to have silent errors when accidentally sending the wrong message.
Why I didn't just use an existing schema packing library:
- Too complicated: I didn't want to have to learn a schema language and format a schema for every object.
- Too slow: I benchmarked a couple of other popular libraries and they were often 10x slower than using the native
JSON.parse. This library is faster than even those native methods.
- Too large: I didn't want to use a behemoth library with tens of thousands of lines of code and many dependencies for something so simple. This library is 400 lines of code with no dependencies.
- Too much overhead: Some of the other libraries that allow you to specify a schema still waste a lot of bytes on padding/keys/etc. I desgined this library to not waste a single byte on anything that isn't your data.
Why not just use gzip compression?
- Bandwidth usage: If you gzip the
playerexample at the top, the payload will actually increase in size. Thus, many engines don't gzip small packets. Compression works best with large payloads with repetition.
- Memory usage: It is common for compression to use an additional 300 kilobytes per connection.
- CPU usage: Per-message-deflate can increase encoding times by 5-10x with small payloads (~2x with large).
- You still can: Using gzip and SchemaPack is not mutually exclusive. You can still use gzip on the array buffers.
These were performed via encoding/decoding the
player object at the start of this page many times with an i7 3770k on Windows 7.
Here's a screencap of the new benchmark.js console output
In addition, SchemaPack really shines when used with large objects with a lot of nesting and long arrays compared to the competition. I encourage you to run the benchmarks with your own objects to see what works best for you.
2.67 KB after minify and gzip without buffer shim.
8.83 KB after minify and gzip with buffer shim.
On the server, you can just copy
schemapack.js in to your project folder and
require it. (Remove the
./ if installed through npm)
var sp = ;
On the client, use webpack/browserify to automatically include the prerequisite
buffer shim if you're not using it already.
For example, if you had a file
index.js with the following:
var sp = ;// More code here using schemapack
You can add the
Buffer shim by typing
browserify index.js > bundle.js and then including that file in your HTML.
Alternatively, just grab the built minified file from the build folder in the Github repository. Then add the following to your HTML page:
schemapack. This built file only needs to be used on the client, as the
node server already includes the prerequisite
Buffer. The server should use the unbundled version.
Build your schema:
var personSchema = sp; // This parses, sorts, validates, flattens, and then saves the resulting schema.
Encode your objects:
var john =name: 'John Smith'age: 32weight: 1885;var buffer = personSchema;console; // <Buffer 20 0a 4a 6f 68 6e 20 53 6d 69 74 68 43 3c 80 00>
Decode your buffers back to objects:
var object = personSchema;console; // John Smithconsole; // 32console; // 188.5
Important array information:
The last item in arrays is both optional and able to be repeated. For example, with this schema:
var schema = sp;
All of the following objects are valid for it:
var obj1 = "numbers": "asdf" ;var obj2 = "numbers": "asdf" 10 ;var obj3 = "numbers": "asdf" 14 7 ;var obj4 = "numbers": "asdf" 0 5 7 ;
The last item can also be an array or object, with any amount of nesting. Here's an example schema:
var schema = sp;
And here's an object that conforms to it:
var obj ="name": "joe" "numbers": -3 2 5 "age": 42"name": "john smith iv" "numbers": "age": 27"name": "bobby" "numbers": -22 1 "age": 6;
Set the encoding used for strings:
'utf8' is the default. If you only need to support English, changing the string encoding to
'ascii' can increase speed. Choose between
Add type aliases:
sp;var builtSchema = sp;var buffer = builtSchema;var object = builtSchema;console; // [ 'dave', 1, 2, 3 ]
By default, validation is enabled. This means that the encode function will include checks to ensure passed objects match the schema.
The build function takes an optional parameter for validation. If set to false, the aforementioned checks will be excluded. Example:
var builtSchema = sp; // Validation checks won't be added to the encode function
To avoid having to pass this flag to each call of build, you can instead call
setValidateByDefault to set the default validation strategy. Example:
Setting the parameter to false will disable validation by default, while true will enable validation by default.
Make single item schemas:
var builtSchema = sp;var buffer = builtSchema;var item = builtSchema;console; // -350
Here is a table of the available data types for use in your schemas:
|Type Name||Aliases||Bytes||Range of Values|
|bool||boolean||1||True or false|
|int8||1||-128 to 127|
|uint8||1||0 to 255|
|int16||2||-32,768 to 32,767|
|uint16||2||0 to 65,535|
|int32||4||-2,147,483,648 to 2,147,483,647|
|uint32||4||0 to 4,294,967,295|
|float32||4||3.4E +/- 38 (7 digits)|
|float64||8||1.7E +/- 308 (15 digits)|
|string||varuint length prefix followed by bytes of each character||Any string|
|varuint||1 byte when 0 to 127
2 bytes when 128 to 16,383
3 bytes when 16,384 to 2,097,151
4 bytes when 2,097,152 to 268,435,455
|0 to 2,147,483,647|
|varint||1 byte when -64 to 63
2 bytes when -8,192 to 8,191
3 bytes when -1,048,576 to 1,048,575
4 bytes when -134,217,728 to 134,217,727
|-1,073,741,824 to 1,073,741,823|
|buffer||varuint length prefix followed by bytes of buffer||Any buffer|
Just clone the repository, run
npm install in the directory to get the testing framework (it also grabs other libraries for the benchmarks)
This library uses
Buffer when in the
node.js environment (always included) and the buffer shim when in the browser (included with browserify/webpack).
The travis tests pass with node versions ranging from 0.11.15 to the latest (6.3.1 at the time of writing).