Protocol buffer versus boost binary serialization

[Pooja Sadevra]

Hi Protobuf team , 

I am wondering if any runtime analysis has been done of protobuf versus boost binary serialization ? I saw a really old thread ( from 2009 ) that said text archive is much slower than protobuf . I am wondering if any updated analysis has been done on this aspect ? Are protocol buffers more runtime efficient when serialising data to binary ?

Thanks

Pooja 

[Em Rauch]

We don't maintain a reference benchmark suite compared to other technologies; unfortunately have also found in general that it is actually very hard to create benchmarks that give results which are representative of real world usage.

Broadly, C++Proto is one of the most optimized binary format parsers in the world though. There's many formats that are faster (including CapnProto and Flatbuffers, but also many FFI binding style libraries or blind struct layout writes) which which are objectively faster despite having much less optimization attention.

Generally it's because in those cases the wire format itself has the characteristic of near-zero cost parses, which comes at the expense of schema evolution ability; where schema evolution ability is generally the top priority for Protobuf because we are intended to be used for long term evolution in RPC systems where you can't update both sides of the communication atomically. Basically, cases where you have limited or no need for schema evolution (because e.g. you are doing FFI within one system where you update both sides) are the cases where other technologies can easily beat Protobuf on performance.

--
You received this message because you are subscribed to the Google Groups "Protocol Buffers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to protobuf+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/protobuf/b75b5a8a-cfbb-4672-9656-feafffb6ee20n%40googlegroups.com.