the reflexions insider

Cloud computing is all the rage these days. Albert Wenger of Union Square Ventures gave an interesting talk at the Web 2.0 Expo in NYC recently which covered some of the implications of cloud computing. At the Web 3.0 Conference there was a lively discussion about some of the more technical issues surrounding cloud computing, software as a service, and web architectures in general. Here are some of the points that were made:

• A general rule of building out any system that needs to scale is to look at every layer in your system as a caching opportunity to minimize processing.
• Start with clusterable technologies wherever possible.
• The cloud has made “spinning up” instances of your application just an API call away. It’s extremely flexible and basically means you’ll (almost) never have to think about hardware again.
• Amazon AWS is the leading cloud solution right now and is extremely flexible because it provides access to the HTTP layer. It’s becoming the Wal-Mart of cloud solutions and it’s price-optimized for CPU cycles.
• IBM could become a major player in cloud computing over the next 2-10 years because of their history of mainframe architecture. The cloud is a lot like a mainframe and IBM understands that model better than anyone.
• Google App Engine was a rush to market. It provides an extremely constrained stack which may meet your needs but if you need variability of any kind, Google simply doesn’t deliver. There’s basically no control over the HTTP layer which is the foundation of the web.
• Remember that in the cloud you lose some control, like getting faster latency between EC2 instances. This can be very frustrating and sometimes must be accounted for.
• Words to the wise: most companies will never need to scale up significantly so you should avoid spending money on this at all costs until it’s necessary. Spending money on scaling up front could kill your startup. This is why the cloud is so appealing. Scalability is built in by design and offers a pay-as-you-go pricing model.

Convinced yet? At Reflexions Data we’re advising our clients to explore cloud computing options cautiously. Every critical system should have a redundant non-cloud instance synched and ready to roll into production at a moment’s notice. (Here’s why.)

Here are some notes from the provacatively titled “Death of the Relational Database” discussion:

• The restrictions that are placed on developers by SQL are largely responsible for much of the bad software that exists in the world. (Glenn McDonald, ITA Software)
• The relational database model isn’t going anywhere soon, but now there are many more options.
• Implementing things like social network analysis and temporal logic are incredibly difficult with the relational model.
• SPARQL as a query language is derived from SQL and therefore inherits much of the “badness” that comes with it. This model forces you to start with the universe in front of you and start looking for matches in a very unnatural way. Sometimes very simple requests require complex joins that are difficult to understand. Humans think in terms of associations. (Glenn McDonald)
• Glenn’s Manifesto: Whole Data:
  1. What logic allows, storage may not constrain (i.e., you must have internal consistency among your triples)
  2. All relationships are lists (the really interesting things in life are multiples — modeling this is the fundamental task of a data modeling and storage systems)
  3. There is no up or down, only onwards (i.e., there is no directionality in data)
  4. Human inquiry follows paths; machine inquiry follows all paths at once.

Stay tuned for the final installation of my Web 3.0 Conference series covering venture capital funding and startups in a changing economic landscape.

You must be logged in to post a comment.