I’ve seen enterprise software with security rules that require cross joins across 3 or 4 tables. This will simply not do for web-scale software. (I’d contend that this kind of rules are potentially so convoluted that only one or two people truly understand how it operates at an organisation)

If I shutdown my instance without first saving its state, then yes, the data is lost. But saving its state and storing it on S3 is a straight forward process. The weakness, of course, is the inability to mount S3 as a virtual drive, but there are several external solutions, including a FUSE-based solution that I am involved with, that help make this process pretty straight forward, so it’s not a problem without remedy.

“There are those out there, such as Microsoft’s Dare Obasanjo, who argues Web scale is far bigger than enterprise scale, and that enterprise architecture practices and skills are largely a waste of time, predicated on complexity for its own sake; that any serious Web company will do better. But anyone that has had the “pleasure” of trying to authenticate on Blogger (”We’re out of beta and ready to go”), or seen the recent downtime problems, knows that Google should perhaps have stuck with perpetual betas. If Web apps are so great what is the problem?

That’s the thing with data integrity- its just not easy, whether you’re talking about identities, transactions, or master data. Sloppy can be good, but its not always good. Sometimes structure is a good thing. The key word is perhaps transactional, which takes us back to database persistence and Amazon Web Services. ”

Perhaps an RDBMS is just not the right solution to deploy in utility compute scenarios. After all, who ever heard of dynamically provisioned databases? How often do we ever move them around between machines?

Could it not be argued that RDBMS’en are linked to the non-web, centralization dominated world of the enterprise data-centre?

Bosworth would seem to think so (at least a little bit):

http://acmqueue.com/modules.php?name=Content&pa=showpage&pid=337&page=5

Maybe we cram far too much into an RDBMS because it’s an easy well-understood option rather than because it’s suitable for the task? One could argue that enterprise should be allowed to continue to do this but it might be that if they wish to exploit these new models they’ll have to give up the RDBMS drug.

It’s not like we don’t have other ways to achieve the equivalent level of transactional integrity that we get with an RDBMS……

By the way, the “big database in the sky” that Marten is talking about (I saw his talk at the Web 2.0 Forum) is about being able to query across multiple web resources. I followed future dialogs with him on this, and I think there is general agreement that what he is describing is the Semantic Web, with RDF and SPARQL. Marten wasn’t that aware of the Semantic Web at the time he proposed the big database in the sky

I think that this is very different from providing a true structured, durable persistence service for a compute node in the compute cloud, which I agree sounds like a *very* interesting opportunity.