26

We have REST webservices that can serve XML or JSON (WCF). I'm toying with idea of implementing Protobufs. Why?

PROS

  1. Less load on servers.
  2. Smaller message size - less traffic.
  3. It is easier to switch now than later.

CONS

  1. Need to be implemented
  2. Going to be harder to troubleshoot/sniff messages for debugging.
  3. I can enable GZip on server and JSON will consume as much traffic

What is your suggestion and/or experience on this?

katit
  • 681
  • 1
  • 7
  • 14
  • 1
    I looked at the wikipedia page, and was actually more chars per key-value pair, so why dose it have smaller messages? – Ramzi Kahil Jun 22 '12 at 16:18
  • Because of serialization. Binary data comes through as binary (byte arrays for example). In addition, it doesn't carry property names in a message. They go by property ordinals. https://developers.google.com/protocol-buffers/docs/encoding#structure – katit Jun 22 '12 at 16:24
  • 10
    Technically you answered your own question, compress the JSON and be done with it. Most importantly you never mention an actual **business** case for spending money and time on this activity. –  Jun 22 '12 at 16:41

2 Answers2

40

Does the business value of implementing them exceed the cost?

If you implement, you need to change not just your server, but all clients (although you can support both formats and only change clients as needed). That will take time and testing, which is a direct cost. And don't underestimate the time taken to really understand protocol buffers (especially the reasons to make field required or optional), and the time taken to integrate the protobuf compiler into your build process.

So does the value exceed that? Are you faced with a choice of "our bandwidth costs are X% of our revenues and we can't support that"? Or even "we need to spend $20,000 to add servers to support JSON"?

Unless you have a pressing business need, your "pros" aren't really pros, just premature optimization.

parsifal
  • 591
  • 5
  • 3
23

i maintain apis and somebody before me added protobuf (because it was "faster"). The only thing faster is RTT because of of smaller payload, and that can be fixed with gzipped JSON.

The part that is distasteful to me is the relative work to maintain protobuf (compared to JSON). I use java so we use Jackson object mapping for JSON. Adding to a response means adding a field to a POJO. But for protobuf I have to modify the .proto file, then update the serialization and deserialization logic which moves the data in/out of the protocol buffers and into POJOs. It's happened more than once that a release has happened where someone added a field and forgot to put in either the serialization or deserialization code for the protocol buffers.

Now that clients have implemented against the protocol buffers, it's almost impossible to get away from.

You can probably guess from this, my advice is don't do it.

Kevin
  • 1,341
  • 8
  • 12
  • 5
    If you're writing (de)serialization code to move data in/out of POJOs, you're doing it wrong. Just use the protobuf-generated classes directly. – Marcelo Cantos Dec 18 '15 at 10:42
  • I think in that case I was moving in/out of POJOs because the codebase supported both JSON, XML, and Protobuf requests/responses, and given that I can't add Jackson and/or JAXB annotations to protobuf generated classes, and I want to use the same model objects throughout the code, I don't know that I have options to just use the generated classes. I can see how this would be possible to do if I was writing say GRPC services that only talk protobuf. – Kevin Sep 10 '18 at 21:27