-
Notifications
You must be signed in to change notification settings - Fork 122
Prior art
In the RSS 2.0 spec, Dave Winer defined a Cloud element, which provides a way for feed publishers to ping the "cloud" when their feed is updated. Before working on the PubSubHubbub spec, we knew about RSS but had never heard of the cloud element; we figure this is because nobody actually uses it.
The cloud element accomplishes the pub part of pub-sub. What's missing is the sub half, which is what makes the whole system useful. Some people have said, "Hey, rssCloud did define subscriptions." In our view the subscription part of rssCloud still just solves the publishing part of the problem. rssCloud only distributes light-weight pings to subscribers-- subscribers still must re-fetch the feed to determine if it's changed. This is not easy for subscribers to do, complicating their implementations, and has scalability issues because of the thundering herd problem.
In contrast, the PubSubHubbub hub is a distinct component that pushes updates to subscribers in response to publisher events. Hubbub pings are "fat" and contain the full Atom or RSS body with only what has changed in the feed. This reduces load on the publisher, lowers end-to-end latency, and simplifies publisher and subscriber implementations. That is what makes PubSubHubbub different: It puts complexity in the center and provides a consistent interface for all three pieces of the solution (pub, sub, and hub). These interfaces are efficient and save bandwidth and CPU time for all parties involved. This is what is needed for Internet-scale publish and subscribe.
So we definitely give Dave credit for coming up with this cloud idea, but we think PubSubHubbub takes this to the next level.
Like the RSS Cloud element, the weblogs.com ping interface and changes.xml file handle the pub part of the problem, but miss the sub half that makes it useful.
These ping interfaces also do not require the pinged URL to the subscribed feed itself-- a channel link is good enough. That means ping server implementors must do auto-discovery on channel links to find the actual Atom and RSS feeds to subscribe to. The extended pinging interface allows you to supply the actual feed URL, but it's optional and big sites don't use it. Auto-discovery is expensive (fetching, parsing) and potentially dangerous because of DoSing by proxy. The PubSubHubbub ping protocol requires the real feed URL so no auto-discovery is required.
Another issue has been that control of these existing pings are currently in the hands of just a few sites. Our goal is to make PubSubHubbub decentralized and free; anybody can run a hub.