Infrastructure neutrality is now an absolute necessity: reflections on the AWS outage
Fact: AWS’ outage last week is the kind of thing that we can expect from data centers forever. Unfortunately, outages are part of the nature of the beast. All hardware infrastructures of all sizes and all degrees of complexity fail sooner or later, and data centers (“cloud” or otherwise) are no exception. It’s what happens next that matters.
Fact: If your PaaS doesn’t allow you to deploy on multiple infrastructures, then the next major data center outage could be as bad or even worse for you and your users as the AWS East outage last week was (or the one in April 2011).
Last week’s Amazon Web Services outage was another reminder of just how essential infrastructure neutrality and portability has become.
In the world of PaaS, infrastructural problems like Amazon’s multiple hours of total downtime have become the elephant in the room, leaving us all with a lot of questions and not a lot of answers from the current slate of dominant PaaS providers (tied to single IaaS providers).
We designed AppFog from day zero as a poly-infrastructural platform and structured it around the principle that we mere mortals can never insulate hardware–even today’s hyper-sophisticated data centers–from the whims of nature and random chance, the so-called “acts of God.”
What we can do, however, is enable you to establish as much independence from error- and decay-prone physical servers as humanly possible.
Freedom from vendor lock-in is not a phrase that we just throw around. It corresponds to a very real possibility that is in the process of realization–and just around the corner at that. In the last week, it has become painfully clear that this freedom is not just a possibility but a necessity.
PaaS solutions have thus far lacked either the imagination or the wherewithal to make infrastructure neutrality and portability a reality. We’re as exhausted as you are by these kinds of scenarios, where you go online to see how your site is doing only to find everything broken, offline, and temporarily–or not so temporarily–dead in the water.
Physical hardware, even data centers built with bleeding-edge technology, will always be faulty. This is why we should never under any circumstances bet all of our chips on a potential single point of infrastructure failure, and this is why we built AppFog the way we did.
The meaning of platform-as-a-service
It’s time to reflect on what a PaaS is and what rightfully deserves the label. Our thoughts on this are very simple: if all that a company is offering is server space in a single data center, they simply do not merit the title platform-as-a-service. This exclusive reliance endangers you, your apps, your ideas, your hard work, and the services you provide.
The whole purpose of PaaS as an actual service is to free you from dealing with purchasing and dealing with hardware resources (virtualized or real). This freedom from hardware is the benefit touted by all the reigning platforms.
But if your PaaS has you locked into a single data center, then you really are still dealing with hardware, just in a way that is more abstract and actually worse than before. Because now you still suffer the often-grave consequences of server failure, except that there’s nothing you can do about it. You were promised freedom from DevOps and what you got instead is a situation in which you can’t do Ops even if you want to. You’re forced to simply wait until things get better.
There is nothing acceptable about this.
What PaaS vendor lock-in looks like
Let’s take one of the current leading PaaS, Heroku, as an example. This is what Heroku looked like last week at the time of the outage:
This is what appfog.com looked like at the same time:
There’s a simple reason for this. Heroku is completely and utterly dependent on AWS East. As goes the Virginia data center, so goes absolutely everything on Heroku. And this remains the case in spite of AWS East having numerous outages in the past. Sadly, Heroku is hardly alone in this kind of relationship of dependence – and this isn’t us bagging on Heroku. Heroku is just the easiest illustration of a problem that many of us are facing.
True, AppFog does in fact support deployment to AWS East. But we also enable deployment through an AWS data center in Ireland. And one in Singapore. And very soon, a Hewlett Packard data center in Las Vegas. And then Microsoft Azure. We have others beyond this already in the works and we’re only in public beta (unlike our leading competitor, which has been around for years).
As we have always said: servers are ephemeral, apps are forever. Let us ask you this: are your apps forever? Or are they still fully subject to inclement weather, inconsistencies in the power grid, and a million other contingencies?