One of the ways in which traditional architectures and deployment models is actually superior (yes, I said superior) to cloud computing is in provisioning.  Before you label me a cloud heretic, let me explain. In traditional deployment models capacity is generally allocated based on anticipated peaks in demand. Because the time to acquire, deploy, and integrate hardware into the network and application infrastructure this process is planned for and well-understood, and the resources required are in place before they are needed. In cloud computing, the benefit is that the time required to acquire those resources is contracted to virtually nothing, making capacity planning much more difficult. The goal is just-in-time provisioning – resources are not provisioned until you are sure you’re going to need them because part of the value proposition of cloud and highly virtualized infrastructure is that you don’t pay for resources until you need them. But it’s very hard to provision just-in-time and sometimes the result will end up being almost-but-not-quite-in-time. Here’s a cute [whale | squirrel | furry animal] to look at until service is restored.

While fans of Twitter’s fail whale are loyal and everyone will likely agree its inception and subsequent use bought Twitter more than a bit of patience with its often times unreliable service, not everyone will be as lucky or have customers as understanding as Twitter. We’d all really rather prefer not to see the Fail Whale, regardless of how endearing he (she? it?) might be.

But we also don’t want to overprovision and potentially end up spending more money than we need to. So how can these two needs be balanced?

The first thing we need to do is know, in a given cloud, how long it will take to provision capacity and put it into the rotation. It would be nice if cloud providers offered a service devops could query to get the “current wait time” (a la customer service queues) but until then this timing will certainly need to be obtained by simply keeping track yourself.

The other “constant” (if there is such a thing in a elastic environment) is the capacity of the instances you are using. We’ll consider this a constant at this point because honestly, we’re not ready to move to the higher levels of enlightenment (and programmability) required to dynamically determine this value – though that will most certainly be the subject of a future, future post. Capacity needs to be in units measurable by the solution aggregating requests (a strategic point of control). This is almost certainly a Load balancer or application delivery controller of some kind, as these components are what enable elastic scalability and basically make cloud work. Typical units might be RPS (requests per second) but because of differences in the way different types of requests consume resources it may be easier and more consistent across applications to use connections, as in “concurrent open connections” as it is one of the limiting factors on capacity of application services.

The other two variables we need are only available at run-time, dynamically. You need to know the existing load – in the same units as capacity – and the current resource consumption rate. The resource consumption rate should be in the same units as capacity and in the same time unit as time to provision. If that’s minutes, use minutes. If that’s seconds, use seconds, and so on. It should be noted that the resource consumption rate is the harder of the two to obtain, requiring access to the historical performance statistics of the aggregating component (the load balancer).

But let’s assume you can and do have all these variables. How is that useful?

The formula is actually a fairly simple one once you’ve got the variables. You’re trying to figure out how much time you have before capacity is depleted (and hoping that the answer is smaller than the time to provision). Just-in-time provisioning, as the term implies, is an attempt to formulaically determine when to start the provisioning process such that capacity always meets demand without over-provisioning. Now, you’re always hedging your bets that a high resource consumption rate will continue in the next “time to provision”. It may be the case that the “spike” is over before the new instance is provisioned, but in this case you’re better safe than sorry, right? Unless your customers like seeing a [whale | squirrel | furry animal] and don’t mind the wait.

Consider the following example:

Total capacity right now is 1000 connections. The existing load is 800 connections. Connections are currently being consumed at a rate of 200 per minute. Provisioning more capacity takes 5 minutes. 1000 – 800 = 200 / 200 per minute = 1 minute of capacity left.

Provisioning should have begun at least 4 minutes ago, and optimally 9 minutes ago (too many years developing software for me – fudge factor included) to ensure capacity was available. In this situation, someone is getting a picture of a [whale | squirrel | furry animal].

The trick for devops is to tune the threshold at which the provisioning process begins. Too soon and you might be wasting resources (and money), too late and you end up with timeouts and angry users. Devops needs a way to programmatically evaluate the results and decide, based on the application (it may be more sensitive to failure than others) and the business significance of the transaction (purchase processes may need more warning than search or general browsing), when it is appropriate to start provisioning in such a way as to ensure availability without incurring a lot of cost overhead.

Now I’m sure this sounds like something out of science fiction, but it’s not. The variables can be obtained, if not easily, and the formula can easily be codified into scripts or management applications that enable this entire process to be automated. At a minimum, it should be possible for any skilled devop (developer or operations focused) to create a script/application/widget/gadget that gathers the data required and displays an alert when it’s time to provision – I suggest a nice HTML interface that encloses the entire page in BLINK tags, because nothing says FIX THIS NOW than BLINKING TEXT, right?

Regardless of how it’s actually to put to use, just-in-time provisioning is the goal of agile operations. How that happens is by leveraging cloud computing and highly virtualized data centers and combining that flexibility with the agility of a dynamic infrastructure. Remember, Infrastructure 2.0 isn’t just about configuration through automation. That’s nice, but it’s not the whole enchilada. It’s also about dynamism and flexibility at run-time, in providing actionable data and capabilities that allow elastic scalability to be truly elastic.