Re: [xml-dev] The Allure of Gothic Markup

[Peter Hunsberger <peter.hunsberger@gmail.com>]

Comments inline below... 

Peter Hunsberger

On Tue, Aug 20, 2013 at 8:53 AM, Simon St.Laurent <simonstl@simonstl.com> wrote:

[...] 

Some of the challenge comes from how we look at scale.  We tend to think of it as something that only works well on identical widgets with identical interfaces.

I suspect - though cannot prove - that scale is actually more about how we direct information processing than about forcing everything to be identical.  Processing things with massively parallel systems doesn't necessary mean that all the systems have to follow the same straight lines.  (I know that's geometrically messy.)

Right now, our models for massively parallel are relatively simple applications that have massive numbers of users.  That's the obvious place to start.  As I watch the costs of these systems drop, though, I think we're going to see many more models emerge.  That's a forecast, not a guarantee, but I definitely see opportunities there.

I'm also curious about how the spread of not only NoSQL but infinitely flexible tools like Git and GitHub change the way we view these kinds of processing.  I suspect much more concrete papers than mine will be talking about the benefits of less regimented systems.

In this regard, the things I watch in the horizontal scalability space for data management are Cassandra, Hbase, Elastic Search, Raik, Reddis, Neo4J and Titan.  There are other similar solutions for most of these, these are just the ones I'm currently biased towards for one reason or the other (like being an Apache Committer).

There are 3 or 4 (depending on how you count them) main data models that pop out of these: big table / columnar data, key value / document stores, and graph databases. Those don't completely displace the relational models, but anyone doing systems architecture that needs to scale in _any_ way (eg. cost) had better understand what they do enable, which can in some cases be 100% of the back end for many applications.  Throw in Hadoop (or similar) and the component mix can provide out of the box solutions for a lot more than data management. The upside is that application designs that properly exploit these components them can take the horizontal scalability for granted, but be aware that the downside is that "properly" is a possibly million dollar issue in itself.  Once put together, this is all far more than scaling the handling of many identical things, among other things it also scales complex analytics both on the Web and displacing the more traditional data warehouse systems. 

Just to bring this full circle, the glue for putting all these pieces together is a whole different topic, but rarely touches any XML related technologies (and when it does that's usually for configuration, not data exchange). However, as I've mentioned previously, I do think that some of the XML technologies could provide some real leverage, in particular, in the graph processing space.

 

[...] 

Thanks,
Simon