News for hackers. Stuff that matters.
Re:The Blink Protocol by badgar on Sunday January 20, 2013 @12:40AM (#5086121)
> Since the blink format contains no metadata, the recipient must have the schema with which the message was written. If schema migrations must be supported, this requires prefixing each batch of messages with the (encoded) schema. This may only be efficient when the batch size is large.

This is essentially a non-starter for me after a couple years of building distributed systems using protocol buffers, for which protocol evolution is the first feature.

Re:The Blink Protocol by jamesaguilar on Sunday January 20, 2013 @02:07AM (#5086261)
They should just assign each schema a unique id, and have the server error out when receiving an ID it doesn't know. Then the client sends over the schema and the server caches it until death. This means the first request bears one additional round-trip, but the average request just carries an extra few bytes of identification.
Re:The Blink Protocol by afuruhed on Sunday January 20, 2013 @02:39AM (#5086313)
Please have a look at the schema exchange specification. We welcome feedback. It is too costly to include meta data in every message. http://blinkprotocol.org/s/BlinkSchemaExchangeSpec-beta1.pdf

Anders