{Warning: The following is a well intentioned rant from an individual who clearly spends too much time thinking about cloud infrastructure services. He would once again like to apologize to his mother that he didn’t go to medical school, and instead chose a remote, cult, fringe sector of the technology industry.}

It was a great week in the clouderati blogosphere, I couldn’t wait till the weekend to re-read and absorb the absurdly important and insightful goodness that hit the intertubes. So here we are, its past midnight on a Friday night, I’ve got a full pint of Pellegrino, I’m wearing sunglasses, hit it

Mr Hoff kicked off a series of blogs about the shouting match brewing between private “cloud” and public “cloud” infrastructure security approaches, with a core theme pivoting around tiered vs. topologically interchangeable network organization. Mr Urquhart wondered aloud about the potentials and limitations of EC2 as a cloud API standard, Ms Macvitte sent smoke signals about the new application world where schema-less data passing of just in time bits replaced the old world of cache ready, HTML content, and Mr Cole sent a simple message in a bottle to the future of the cloud, more features please

Watching every tweet, blog, forum post and customer story coming out of the cloud every day for years is a bit like watching a time lapsed ‘picture a day’ face study. Sometimes you see patterns, but as the fractal unfolds over the fullness of time the information turns out to be the jitter of perspectives more than a core transformation. 

The themes this week, however, were not jitter–they are the cutting edge of the “cloud” evolution. The cutting edge of cloud has shifted from an “if” to a how, and the minds who saw the cloud coming years before the general public are now confronting the balancing act between intrinsically lower featured, topologically insensitive cloud formations represented by certain volume leading de facto standards–and the increasingly technically possible emergence of more topologically calibrated and therefore feature rich clouds. Because simple infrastructure clouds came first, they have largely defined the market–but the voices of “no thanks!” are too powerful and persistent to write off purely as technological laggards–there are important architectural reasons they have abandoned or stayed home so far. 

Instead of devolving my Friday night fun into an armed shouting match about discerning customers dissatisfaction with current offers, I wanted to try on a thought model that’s been stuck in my mind all week. (I’ve tested with no less a cloud expert than Mr. Swidler himself)

The constant cries of “commodity” in the cloud really revolve around a unit of CPU, memory and operating system. If your application is executable within one of these modest entities, ideally running in parallel on another for availability because the platform doesn’t guarantee it, the cloud of today works pretty well. 

I call this the slime-mold model. Slime mold is a relatively unique protist, existing as a single cell organism until an abundance of resources appear, when they release a signalling chemical, and join into a larger but very basic organism. None of the cells are specialized, and the cytoplasm around them gives them only a very sparse inter connectivity and protection. 

Yet when uber application innovators such as Facebook talk about building complex applications cellular specialization is very important–I call this the neuron model. Even Google, known for only using commodity CPU/memory combinations organizes them not in vast data-center seas, but in 250 CPU switched clusters for performance and manageability. Because requests on Google or Facebook require hundreds of server to server messages each, the network locality of these servers, their performance and precise arrangement and configuration matter. Mr. Gourlay in fact suggests that in today’s data centers 90% of all traffic is server to server. Corporate data-centers today, where much of the IT spend occurs, have even greater needs and requirements for specialized organization, tuning and configuration. In the thick cytoplasm of some public clouds today, there is very little QOS or SLA on the performance of that server to server traffic. 

The neuron is the ultimate example of cellular specialization. Slime mold doesn’t have any–consuming simple food through massive parallelism doesn’t require it, but if you are human there are 100 trillion specialized synaptic connections between your neurons. These are made possible only by a complex approach to organization–while your core cells are similar, one might say commodity, the way in which your DNA (software) allows them to organize, specialize and connect is anything but a commodity across biological existence. 

While the model is imperfect, it fits many of the themes from this last week. It also informs the future of the cloud. Will all computer applications become highly suitable for a slime mold model cloud? Or will the ability to create network constructs, compute in tight proximity, and future features we haven’t dreamed up yet allow those currently on the sidelines to join the fun. 

Right now as is evidenced from Mr. Hoff’s post above, the two sides are shouting at each other from the extremes. I for one hope we get the ability to form more complex topologies in public clouds very soon. 







One thought on “The Unfolding Cloud Fractal: Of Slime Mold and Neurons

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s