Brett Slatkin wrote: > On Sun, Oct 18, 2009 at 12:50 AM, Tim Bray <[email protected]> wrote: > >> Seems to me that the client model for processing a single vs. >> aggregated distribution might be quite a bit different. And also, the >> original upstream feed might have used entry/source already (this >> makes me nervous about the whole notion of PuSH co-opting <source> for >> its own purposes). >> >> I was wondering if you might want to put an extension element here as >> a child of feed, before the entries start, in a pubsubhubbub >> namespace, saying "the following are aggregated by the hub". You can >> do this safely because Atom has MustIgnore on markup it doesn't >> recognize (hint hint). >> > > This is the first time I've heard someone point this out. I believe > the atom:source element was specifically included in that spec for the > purpose that PubSubHubbub is using it. Bob Wyman seemed to indicate > the same thing too in some other email threads on this list. Could you > clarify how this is "co-opting" the source element? > > >> [Um, when I read this section, there's a little voice in the back of >> my head shouting "YAGNI!"] >> > > I disagree with "YAGNI" here. Take world-wide RSS traffic. Multiply by > 1,000,000. We will need aggregated delivery to fully utilize links. > > -Brett > I can't really judge either way about YAGNI, but certainly IAGNI in this case. :) For any subscriber not getting multiple updates per minute (because they're only subscribed to hundreds of feeds, say, and not tens of thousands), aggregated delivery introduces some complexity to parsing, and doesn't really give any benefits.
It really depends on how PuSH is used in practice. If the majority of the bandwidth is used by a few clients subscribing to many feeds, then aggregated delivery could be really important. If the majority of bandwidth is going to a lot of clients, each subscribing to few feeds, then it really won't be. I personally think that once personal news aggregators support PuSH we'll see a huge shift in the latter direction, but that's really just guessing on my part. I still think aggregated distribution should be made optional, and controlled by an extra parameter when subscribing. --Ravi
