----- Original Message ----- From: "Ted Lemon" <Ted.Lemon@xxxxxxxxxxx> To: "Dean Bogdanovic" <deanb@xxxxxxxxxxx> Cc: "IETF-Discussion list" <ietf@xxxxxxxx> Sent: Tuesday, December 02, 2014 2:59 AM Subject: Re: Blog: YANG Really Takes Off in the Industry On Dec 1, 2014, at 8:54 PM, Dean Bogdanovic <deanb@xxxxxxxxxxx> wrote: > this is one part I don't understand. Why adding another language would make them less agile? If the yang model isn't a good representation of what is being modeled, it can cause more harm than good. Same problem exists with MIBs. <tp> The difference is that MIBs are written in a much simpler language; what object should there be, its syntax and status (index, read-only, read-write) and that's about it (even augmentation tends to confuse many people). I have never yet met a MIB module that I could not reverse engineer to determine the design, even the requirements. YANG is different, it is capable of much more complicated things and occasionally it is unclear what it does mean (something that surfaces on the netmod WG list now and again). What I think I see happening is what happened when programming languages became more widely used, an inability to keep things as simple as they could be, resulting in code whose purpose was unclear, that was error-prone and hard to understand or maintain. Not an issue with SMI. The YANG models of the IETF seem to be diving into complex code from which it is hard to discern what the purpose is, and the fact that most of the exemplars are written by those highly expert in YANG, and so use the wide range of constructs available, does not help. Perhaps the IESG should require that any IETF YANG data model must be accompanied by an information model so that we can debate what should be done independently of deciding how to do it. After all, this is a stricture that has been imposed on the I2RS WG. Tom Petch When different implementations of the same thing use different base assumptions, it can be difficult to come up with a management model that is congruent with all of the different base assumptions and is still useful. I wouldn't say it's impossible, but it's a good bet that a poorly thought out model or a model that is based on experience with a single implementation will fail in this regard.