DAG-PB is a codec that implements a very small subset of the IPLD Data Model in a particular set Protobuf messages.
DAG-PB can be parsed by Protobuf libraries, and can be emitted by Protobuf libraries (with some caveats -- DAG-PB is actually stricter than Protobuf in some details).
DAG-PB implements only a very small fraction of the IPLD Data Model. You should not expect to be able to take arbitrary IPLD documents and emit them as DAG-PB, nor to be able to take documents from other codecs (e.g. DAG-CBOR) and transliterate them into DAG-PB -- DAG-PB isn't that flexible.
DAG-PB is mostly seen in IPFS, where it is a backbone of how Unixfsv1 data is serialized.
See the DAG-PB specifications for details.
DAG-PB is based on Protocol Buffers, and Protobuf 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 write DAG-PB interactions using other Protobuf libraries, although it might require some care.
DAG-PB is a subset of Protocol Buffers: It is a specific schema within Protobuf, and also imposes some constraints that not all Protobuf systems will enforce. Therefore it will be necessary to exercise some caution if making your own DAG-PB interactions using other Protobuf libraries.
(Standard performance caveat: implementation details always matter; performance is always situational; always do benchmarks before presuming serialization performance is a critical bottleneck; etc.)
DAG-PB 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-PB 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-PB by hand.
DAG-PB is not general. Only very specific structures of data can be encoded in DAG-PB. It is very unlikely that unstructured data in the Data Model will be encodable in DAG-PB unless it has exactly the known structure matching DAG-PB.
See DAG-PB Specs: Logical Format for details on the specific structures which are encodable in DAG-PB.
DAG-PB is slightly higher density compared to the same data encoded in DAG-JSON or DAG-CBOR, because it does not encode the keys for its primary structure, which would otherwise often be redundant.
However, the tradeoff for this is DAG-PB's lack of generality, and lack of self-description. The same design choices which give DAG-PB this density inherently mean it can be used for a narrower range of data structures.
This key elision also does not apply to any of the user's content stored within DAG-PB. DAG-PB does not have any compression features.