Document: draft-ietf-ipoib-architecture-03 Reviewer: Michael Patton Date: March 30, 2004 Summary: This draft is basically ready for publication, but has nits that should be fixed before publication. This document is pretty good, but is somewhat schizophrenic on what its purpose is. Section 1 (which is half the document :-) tries to be a summary of what the IB architecture is, as a basis for the later parts that describe how IP runs over IB. But then some of those later parts seem to say that IB needs to have certain additional features before IP can be properly implemented, thus being recommendations to the IBTA. The document has a lot more detail on how IB works than on how IPoIB works. If the document were undergoing another major revision, I'd suggest making section 1 a little terser, and expand a bit on some of the terseness in the later sections. There also seem to be a few things left open (like how Q_keys [I think] get picked) which may cause interoperability problems. But this is an architecture document, not the actual spec, so I think it's probably OK even with that. Herewith the nits. All can be easily fixed, or at least it seems so to me. References in Abstract References to RFC1700 which is obsolete Sec 1.1 has a normative reference to a URL, I thought that wasn't allowed. In any case, it should be a real reference in the references section. The term "nibble" used but not defined (I suggest below in correcting a "typo" to change it to "4 bits") 1.2.1.1b Isn't Site Local now deprecated? That should be noted. I'm not up on IPv6 address architecture so can't comment further on what it says about IPv6 addresses. At one point it mentions the use of "Traps 66 and 67" but doesn't talk about how these traps work. I realize that the details are in the actual spec, but the document should probably say at least a few words on this. References not divided into normative/informative, but I think all are normative. The body of the individual references is just doc titles, references should have more than that. Section 2.0 has the phrase "MIBs ... in separate documents" but no reference. The way the paragraph is written, it seems these missing references are normative (and thus their absence is a fatal flaw). In looking at the drafts directory there do seem to be MIB drafts, they should be referenced. Hitting section heading 2.0 halfway through the document suggests that it could use a bit of work on structure, but that's no reason to slow it down. ================================================================ Typos: 1.2 "There are a two exceptions." Remove the middle word. Several places in 1.2 "across an IB subnet" really means "across an IB router", or "between two IB subnets". I think I saw this odd word usage in other places, too. 1.2.1.2 "The nibble, denoted by x above" that comma doesn't belong there. 1.2.1.2 "The 4 bits, denoted by y in the GID above" that comma doesn't belong there. Also, those two phrases should be ade more similar, either "nibble" or "4 bits" in both places, but if "nibble" is used it needs to be defined. I suggest adding "in the GID" to the first to make them the same. 4.0 "parameters such as, Q_Key, Path MTU, needed" the first comma doesn't belong there, the second comma should be replaced with " and". 1.3.2.4 "The SA may be requested by the ports to generate" would read better as "Ports may request the SA to generate" 4.2 the diagram has a spacing error.