Universal Binary JSON

UBJSON Running Example

(size-optimized UBJSON)

[object-begin, 1B]
   [small-int, 1B] [str-len, 1B][title, 5B] [string-field, 1B]
                        [small-int, 1B] [str-len, 1B][Back to the Future, 18B]
   [small-int, 1B] [str-len, 1B][sub-title, 9B] [null-field, 1B]
   [small-int, 1B] [str-len, 1B][year, 4B] [short-integer-field, 1B][1985, 2B]
   [small-int, 1B] [str-len, 1B][imdb-rating, 11B] [float-field, 1B][8.5, 4B]
   [small-int, 1B] [str-len, 1B][keywords, 8B] 
      [array-begin, 1B] [strong-type-marker, 1B] [string-type, 1B]
         [small-int, 1B] [str-len, 1B][time travel, 11B] 
         [small-int, 1B] [str-len, 1B][delorean, 8B]       
         [small-int, 1B] [str-len, 1B][comedy, 6B]            
      [array-end, 1B]
   [small-int, 1B] [str-len, 1B][release-dates,13B] 
      [array-begin, 1B] [count-marker, 1B] [num-elems, 1B] [strong-type-marker, 1B]
         [1985, 2B] 
         [1986, 2B] 
         [1987, 2B]
         [1992, 2B] 
         [2008, 2B] 
         [2010, 2B]
         [2012, 2B]
         [2015, 2B]
         [2016, 2B]
[object-end, 1B]  

155 byte

Comparison to Carbon

Design Goals



  • UBJSON is not intented for writes, while Carbon emphasises in-place modifications including (optional) reserved memory ranges as capacity buffers to avoid reallocation during updates
  • No specification on update semantics for UBJSON, while there is a detailed specification on atomic operations, update behavior and version control for Carbon



  • Carbon and UBJSON are very similar in its core structure (similar containers in general), similar use of markers.
  • Support of constrainted arrays ("strongly typed arrays"), but with different optimization goals in mind


In direct comparison, Carbon focus more on semantics (e.g., embedding additional meta-data) and focus more on support for operations of typical operational document stores (e.g., on-the-fly updates), while UBJSON focusses on efficient data exchange between two actors. Both, Carbon and UBJSON share the concept of a minimalistic type-support to have full compatibility to the latest JSON specification. However, Carbon adds one additional data type (custom) binary to enable user-defined function (UDF) support on user-defined types, though.

  • Carbon supports three-valued logic for boolean values, and encodes null values not only for the JSON constant null but also as valid value in strongly typed arrays (e.g., intermixing of null and number values in column containers of Carbon). UBJSON cannot store a null value in a strongly typed array of a particular non-null type, and must fallback to size-inefficient standard arrays in UBJSON.
  • Variable-length integer encoding for string lengths, or counters in Carbon
  • Encoding of binary data: while in UBJON a (strongly typed) unsigned 8-bit integer array is used, in Carbon there is a dedicated type binary (resp. custom binary) that encoded the actual binary string data type (i.e., the what is stored information gets not lost in Carbon as it does in UBJSON)
  • Revision numbering as dedicated (variable-length) meta-data field in Carbon
  • Record identification via (optional) primary key as meta-data field in Carbon, and defined semantic about record identity
  • Overall, typically smaller binary size for Carbon files compared to UBJSON files
  • For Carbon, the evaluation of dot-path expressions is well-defined and specified
  • Carbon specificies two way to express absence of data, null and undef where the first one maps to user-defined data and the second is a dedicated result value for dot-path expressions.