DAG-CBOR is a codec that implements the IPLD Data Model as a subset of CBOR, plus some additional constraints for hash consistent representations. DAG-CBOR also adds a "link" type using a CBOR tag, to bring it in line with the IPLD Data Model.
DAG-CBOR is a codec we recommend using. It has very complete mapping to the IPLD Data Model; it's binary and length-delimited (meaning it's generally very effecient to parse); and it's widely supported through the IPLD ecosystem.
See the DAG-CBOR specifications for details.
DAG-CBOR is based on CBOR, and CBOR libraries are already ubiqituous and can be found in many languages. If you are working in a language that doesn't have IPLD libraries yet, you can use another CBOR library easily.
(Standard performance caveat: implementation details always matter; performance is always situational; always do benchmarks before presuming serialization performance is a critical bottleneck; etc.)
DAG-CBOR is typically considered to be fast. It's a binary, length-prefixed format. These traits usually associate with good performance.
(In length-prefixed formats, strings can be scanned quickly without escaping, and encoded without escaping; allocations can be made up front; etc; all these things contribute to speed and efficiency.)
DAG-CBOR is not very human friendly. It's a binary, length-prefixed format. While these traits contribute to its performance, they do not make it easy to edit.
You cannot typically write or edit DAG-CBOR by hand.
DAG-CBOR supports the full IPLD Data Model, completely losslessly.
DAG-CBOR's density is roughly comparable to JSON. E.g. If a series of maps have the same keys repeated, then those keys will be repeated in the DAG-CBOR serial form.