The Future of Infrastructure is … ‘Optimal’
I’ve had many conversations and read multiple articles in the past few months in which people try to predict what the future of technology infrastructure will look like. I’m personally excited about this as it brings to light the importance of a well-planned infrastructure, and it also raises infrastructure awareness amongst everyone from the product and marketing teams to the C-Level executives. However, I feel that the prediction of “hybrid,” as well as the overuse of buzzwords, has left us with a wide gap of missing details and context. Over the next few weeks I’d like to dig into the different types of major infrastructure categories and illustrate how you can find your Optimal Infrastructure Profile by using data and available tools.
Categories of Infrastructure
Let me be one of the first in the room to admit that I do not know exactly what the future of infrastructure looks like. We can all make some fact-based estimations, but we cannot know what the future holds - not simply because technologies are evolving so rapidly (underwater servers!), but also because there has never been (and never will be) a single, one-size fits all infrastructure solution. Lets start by taking a stab at the major categories. The deployment types below have the ability to serve an organization’s complete set of traffic and applications needs today and don’t include PaaS or technologies that address only distinct pieces of the infrastructure puzzle. In no particular order besides number 1:
1. Amazon Stack - In our lifetimes we will rarely see a product and technology affect and shift mentalities, strategies, and ideas about web infrastructure as much as Amazon’s AWS. I’m impressed on a daily basis with what they’ve accomplished. ‘Cloud’ has been around for much longer than most people think, but Amazon has become the ‘Kleenex’ of the public cloud world.
2. Other Public Clouds - Names that commonly pop up here are Joyent, SoftLayer, Azure (although I’ve actually never met anyone using Azure …. dead serious), and Rackspace. Pricing and quality differ but I generally feel that no one has yet to significantly differentiate themselves from the pack - although Joyent is the most interesting to me due to SmartOS. Its hard to argue that Rackspace isn’t the clear number two (they are from a customer perspective I believe), but this mostly comes from their massive, existing managed hosting base, not necessarily from product differentiation. This brings us to number three.
3. Managed & Dedicated Hosting - Rackspace has dominated this space and there are too many other players to mention in one post. The idea is simple: you get a fixed, non-shared server resource (usually) with very little to no control over the networking layer (local switching, VLANs, firewall, and routing/upstream) and require assistance from tech support to do things that otherwise could be seamlessly accomplished within a cloud environment.
4. Own and Operate / Buy and Build - The original, tried and true method of leasing datacenter space, buying (or leasing) hardware, purchasing IP transit, peering with networks and managing your own network and servers through a variety of tools (mostly open source). This market is significantly bigger and growing much more quickly than most writers, bloggers, and investors believe, and at the end of the day, this is how the cloud providers themselves lay the groundwork for their own IaaS offerings.
5. CDN - I’m including content delivery networks here, not only because most CDN’s will give you soup-to-nuts offerings to store, upload, serve, and monitor your entire web presence, but also because I just love CDNs. I ran operations for one and have studied performance and used many more since 2005. Some of the absolute best and brightest minds in the infrastructure and performance worlds have worked for CDN’s over the past few years, and this is still true today. I also think that the newer (newer than static object caching and video delivery that is) products such as ‘application delivery’, code execution on the edge (really just a cloud-based PaaS offering in this capacity), and flexible DNS tools are infusing excitement and new revenue into the content delivery market. Cotendo’s (very profitable) exit to Akamai is a perfect example. However, there are still several underlying obstacles to long-term profitability in the CDN market compared to public cloud products: differentiation is even more difficult than public cloud, Akamai is the ‘AWS of CDNs,’ switching providers is ridiculously simple, margins are scary low, and your addressable market is primarily companies that own their own servers or managed hosting companies who white label. Amazon has made Cloudfront (which is quite the pedestrian product) so easily integrated to an already massive customer base that traditional CDN’s have a hard time selling into AWS customers. Its important to note that almost every CDN wants to - and is trying to - create a viable, long-term cloud product to compete with AWS (Fastly - which I love - may be an exception here as it seems they are focused on CDN performance and flexibility).
6. Private Cloud – This is one of the more interesting ‘technologies’ out there, as the term is massively overused. I have also yet to see a good, consistent definition of a private cloud. Don’t get me wrong, I think out of the six options listed here, the private cloud has the highest growth potential, but the fight over what cloud software wins (hey there Openstack/CloudStack/Eucalyptus/vmWare) creates more noise than signal. For now (and I could be wrong) my feeling is that a private cloud must consist of the following seven components:
- Completely isolated and transparent server iron
- Completely isolated and transparent Layer-2 switching environment
- Complete transparency over where your servers live (address, room, row)
- Complete transparency into first-hop routing (your upstreams/peers or your providers’ if its not in your DC)
- A virtualization capable / ‘cloud-aware’ OS (lets leave this simple and broad for now)
- API’s and ideally a UI that give users automated provision/destroy/bill/etc capabilities
- You must use Arista switches (just kidding, but it does seem quite popular with the private cloud crowd)
There are many folks more knowledgeable than me in this arena, and there will likely be disagreement on what components a private cloud must have. However, this is at least a start, and defining this will be important to the mainstream adoption of private clouds (note that I didn’t use the word ‘standard’ anywhere here).
I should reiterate that there are many other *aaS products that make up an infrastructure profile or serve specific applications that are not included here: DynDNS (a big fave of mine), PaaS products such as Heroku, etc. However, I believe the above list covers the majority of the technologies that can serve, by themselves, an organization’s entire production infrastructure.
Next week we will get into historical preferences of infrastructure and start to talk about how to figure out what the right infrastructure profile is for your company.
- Jason Evans, Co-founder, Stackpop
Follow @jasonhevans on Twitter
Email: jason at stackpop d0t com
Sign up for a free Stackpop account at http://stackpop.com