protocol buffers has implementations in multiple languages and is a stand alone library make it attractive from a supporting other languages standpoint and adoption. And it's licensed under the apache license so hadoop can include their code.
I haven't seen anything technically better about them than Record IO - just adoption and interoperability. I do think thrift has a lot of advantages in the way it has decomposed things but, variable sized ints are not one of them. -- pete On 9/9/08 9:15 PM, "Milind A Bhandarkar" <[EMAIL PROTECTED]> wrote: > Hadoop has had var ints for the last two years under apache license (part of > hadoop record IO), so any other scheme's license is not an issue here. Is > there some inherent goodness in protocol buffers varsize ints that we should > adopt ?
