Virtual Fish are bigger than the Superbowl

Cafe’ world, Zynga’s fastest growing game. It is estimated to have
over 100 million unique active visitors per month. In ad market terms, that is 1/3 the US population. Compare that to the superbowl (nearly 100 million according to neilson).

One of the interesting facts that I took away from the virtual goods summit is the fact that zynga sees 10 dollar a year ARPU (Average revenue per active user over 1 year).

I know, according to all the VCs, that the advertising revenue model is dead, but seriously, placing an ad on these facebook games is dirt cheap compared to the superbowl. AND the quality of the data you can get from a social media campaign is tremendous.

It takes an abused metaphor to make something grow huge, fast

It takes a village to. We have all heard that cliche. Growth like what we see in the social media market needs support. Need creates demand, demand typically filled by meta industries that sorround and support the primes (Prime content developers, zynga, playdom, playfish). The social media meta industry is blossoming: Monitization infrastructure is now an industry (we’ve known that for a while, since we have been working with vindicia, aria, etc. etc.). Metrics companies that compete with cognos are sprouting, as we know there is a huge gap in the promise of cloud and cloud like infrastructure. For once, development tools are keeping pace, mostly because the state of the art is very simple apps, built mostly around flash, and the very very simple facebook API.

Marketing is not my thing, making stuff work IS

That simplicity does not yet extend to the back end technologies that run most social media networks. The LAMMP stack is not keeping up(Linux Apache MySQL MemCacheD PhP). It is a technology base, and architecture, that has not changed much in the last 15 years or so. It is  based on an architectural pattern that dates back to the late 70s(I am referring to MVC). There have been enhancements,  but no fundamental changes to the pattern.   load balancing, concurrency, and scale out are not natural to the pattern.

I would rather have pie

Facebook’s architecture is a step toward the ideal, squishing the technology stack down to only 3 basic layers,  but the architectural ideal would be to flatten the architecture down to one single layer. Rather than web sites being modeled after a layered cake, ideally they would be modeled after a pie (mmmm pie . A single amalgam of infrastructure, that can grow “out” as much as needed. insert mouth watering allegory to pie metaphor here.

Personally, I think we are close. One of the technical reasons for multi layered approach to web site design is the capabilities of each individual component(server). Another reason (or side benefit, take your pick) is security.

Why we have cake instead of Pie

The Multi layered architectural model is a result of economics: Expensive computers are optimized for various workloads. Basic economic theory, when a resource is scarce, it will be optimized for it’s largest most profitable market. This is the driving principle behind server design for the last 50 years. Computing power has been scarce, so the computers themselves have been optimized for specific workloads, like running databases. In the late 90s we saw the rise of generalized high density servers, and web server software, dominating the web tier, then a few years later, dominating the app tier. Now we are seeing these generalized high density servers becoming capable of running databases. So, finally, the same high density, low cost, servers are physically capable of running all the software systems used by high scale internet web sites. Which is a step. We are only a generation, or so, away from having a single physical server have enough capacity to run the entire stack of software for a high volume web site. For small websites, we are already there. Most word press based blogs use the same basic architectural pattern as a huge social media web site(web server, app server, database).

The management overhead for running the entire software stack is still too much for a single commodity server. Specifically VMWare and Zen combined with the OS still takes too much compute power away from the application. High volume web sites are still running their entire server farm at very high utilization rates.

Tipping point ahead?

Two things can/will happen.

1. The next couple CPU generations will begin to create the kind of compute power that web sites require in order to completely flatten the physical hardware stack.
2. The movement  of the virtualization functionality down into the hardware (arguably where it belongs). Ideally there should be code in the firmware of the chipset on the server motherboard that provides much of the low level functionality now provided by VMWare/Zen.

The Physical ideal

The ideal is this: A single, low cost, high density server with enough CPU cores, Memory and memory bandwidth to run OS instances for the web server, app server, and Database, as well as run in-memory IP networks and firewalls between the layers. AND provide lower latentcy and higher concurrency than individual systems connected by networks.

Once this physical ideal is reached, then we can develop enough experience to create a new software pattern that more naturally fits network application development.

So what, Now and Later?

So where to invest, where will contributions have the most impact, where are the control points?

In the short term, Anyone working on low level virtualization technologies. nVidia could make a short term killing by coming up with a chipset that could allow Zen/VMWare to virtualize a system with GPUs. (yes GPUs, a surprise technology in server side computing, when physics and stream based processing become important to increasing concurrency, and reducing latency). Also look to the industry veterans, Intel and AMD.

Long term investment, anyone doing research into a new architectural pattern that will replace MVC.  MVC has been in use for nearly 30 years now. The next architectural pattern will be developed in a fraction of the time, and because of the fundamental change in the economics of computing, it’s adoption rate will be unlike any seen previously.

It all comes narrowly back

This accelerated adoption cycle will dramatically narrow the number of players in the meta industry (and you thought I was just rambling, and that thread went nowhere). Put bluntly, when the architecture and techology replacing MVC comes along, it may be only come from one organization. With a likely adoption rate so accelerated that competition has no time to develop.

Interesting times ahead

Imagine 90% of the web traffic on the internet switching from one architecture to another in under a year. If zynga can develop and deploy an app in 5 weeks, that reaches 50 million people in 3 weeks. As the farmer used to tell me “that tells ya somethin right ther”.