World Wide Web Without Walls Micah
Brodsky, Maxwell Krohn, Robert
Morris, Michael Walfish, and Alexander
Summary: Everything has moved to the Web, but unfortunately users data
is spread all over, across different sites and portals. This has
drawbacks (i) it makes it difficult to share data among the
applicaitons. for this the websites implement API and
This works but the applications need to know each others APIs. (ii),
since the data lives with the applications, it is difficult to
migrate. (iii) and applications have control over the data,
of the user itself.
We propose a new architecture where a shared server provider that
provides storage, network, and access control, and provides APIs that
applications can use. This architecture will give more control to
users. All the applications can access the same data from the same
users. Another advantage is that new applications can immediately
access the users data.
Challenges include, applications may be buggy, why would anyone want
to use this framework, and what is the incentive for the storage
provider. Yip showed how these are addressed and showed that W5
natural ways of addressing these concerns.
Q & A:
Q: You might have something of a bootstraping problem, the
reason that places like FB have a lot of data is that they have a
A: To bootstrap we will need a few small applications, that use
Q: I presume that they will be multiple providers, and then would
the data again be fragmented.
A: We would like to have multiple providers, but only thing that
needs to be done expanding the security ring.
Q: What about data format.
A: Today we have open software that can access various formats.
Comment: But that is after a lot of arm-twisting and time.
Q: Would it cause additional latency and performance hit.
A: We assume that there would be a cluster and data would be directly
accessed from there, but if needs to be copied over the network, then
there would be latencies.
Q: What would be the impact on the programming model.
A: We have a small prototype, hopefully there iwll not be a major
Q: What is the scope. Are you trying to convince Flickr and
use same data. or are you more focussed on smaller players.
A. We are not suggesting that existing applications move to this model,
advocate that the newer applications should use this.
Q: What fraction of users are activily using multiple sites, and
A: Many users use multiple services.
Q: How do you deal with applications that might corrupt and modify the
A: That is true, we will need some type of reputation of application.
You will have a filesystem that has roll-back support to undo such
Q: How many of the problems that a typical system encounters would you
need to face.
A: Yes we will have to take care of that.
Q: It seems like a good match with Allman's NIDs
A: Yes there are multiple ways of doing this. One way that you
this is that users have virtual machines.
Q: Why would users contribute applications if there is no money to be
A: I think that many people are willing to write open source free
Q: The permissions/policies can be difficult, and do you expect
users specify these and get them right.
A: It is different from Facebook, in that here you have to make this
decision once per resource, rather than all the applications, so in
that it could be simpler.