JLD2 supports compression of isbits arrays. This includes the typical Array{Float64} but also arrays of custom structs that are immutable and only consist of basic number type fields.

To enable the default compression, you can write"example.jld2", "large_array", zeros(10000); compress = true)

using the FileIO API. Alternatively use

jldsave("example.jld2", true; large_array=zeros(10000))


jldopen("example.jld2", "w"; compress = true) do f
    f["large_array"] = zeros(10000)

When reading a file JLD2 detects compression and automatically decompresses the data so it is not necessary to pass any extra parameters for that case.

JLD2 uses TranscodingStreams.jl to interact with compression algorithms. Compression is disabled by default and if compress = true is passed, CodecZlib.jl will be used.


Note that CodecZlib.jl must be explicitly installed for compression/decompression to work. If the package is not already loaded JLD2 will try to dynamically load it, but it is recommended to do it explicitly, e.g. using JLD2, CodecZlib.

Choosing a compression algorithm

If you want to use a different compression algorithm that is better suited to your needs, you can also directly pass a compressor.

CodecZlib.jlZlibCompressorThe default as it is very widely used.
CodecBzip2.jlBzip2CompressorCan often times be faster
CodecLz4.jlLZ4FrameCompressorFast, but not compatible to the LZ4 shipped by HDF5

To use any of these, replace the compress = true argument with an instance of the compressor, e.g.

using JLD2, CodecBzip2
jldopen("example.jld2", "w"; compress = Bzip2Compressor()) do f
    f["large_array"] = zeros(10000)

Manually selecting compression for datasets

Sometimes you may know, that some of your arrays are easily compressible and that for others it is not worth the effort. For precise control, the write function takes an optional keyword argument to override the file compression settings.

using JLD2
jldopen("example.jld2", "w") do f
    # This can be efficiently compressed  → use compression
    write(f, "compressed_array", zeros(10000); compress=true)

    # Don't compress this 
    write(f, "large_array", rand(10000))