Here is an aspect of the behavior of your future business applications that you may never have had to consider before: Suppose as you fully transition your services to cloud platforms, you move to this more efficient model of microservices delivered through this emerging format called containers, championed by Docker.
Since your services will be scaling up and down with customer demand, will your IT department or DevOps professionals require behavioral tracking to keep tabs on the performance of your applications?
There’s a cottage industry emerging around utilizing the same analytics functions that track the behavior of your customers, to track the behavior of your applications.
“Fundamentally for an operations team, it’s another level of complexity,” said Sahir Azam, who directs product management for a company called Sumo Logic, “because there’s another layer of abstraction.”
The first wave of virtualization set some precedent, especially for VMware, which most effectively capitalized on the increased complexity of data center management brought about by the head-on impact of virtualization with cloud platforms.
Organizations such as Docker Inc., which successfully concluded its second annual DockerCon in San Francisco this week, have argued that their second wave of virtualization — containerization — brings about a radical simplification of this complexity.
But on the other side of that argument are organizations that hope to replicate VMware’s success. On Monday and Tuesday, Sumo Logic and several others, including VMware itself, made the case that microservices architectures will bring with them yet another layer of abstraction, called micro-segmentation.
And with that new layer, they say, comes the need for oversight.
“Virtual machines were harder to manage,” argued Sumo’s Azam in the midst of demonstrating Sumo Analytics, “Because the rate of change was higher. Now I think of Docker as taking that up to another level, where you spin up hundreds of thousands of containers in a few seconds, to scale your app or run test analytics — which is awesome, and which is why people are looking at these platforms.
“But that’s generating a vast volume of data,” Azam continued, “that’s hard to analyze with traditional tools. And just the architecture of new apps that people are writing on these containers, is highly distributed. You had monolithic applications, then you had client/server, and now you’ve got microservices architectures, which have a complete web of interaction. We think that analytics is the only way to solve that challenge.”
Real Time Data
Analytics services such as Sumo Analytics could potentially give CIOs, department heads, and people in charge of DevOps and development teams who are not necessarily familiar with the process of software development itself, oversight over the performance and connectivity of their customer-facing applications in real time.
Literally, if and when something goes amiss, an alert would go out through the browser-based portal that could be picked up via smartphone or tablet.
Azam demonstrated one example of how a team manager may be able to drill down through the Sumo Analytics dashboard to get real-time insight. Using the same algorithms that make customer analytics work with big data, Sumo can plow through, say, all the logs for customer-facing servers over a given interval, such as the last five minutes.
Even in Sumo’s live test environment, which was only generating simulated data, just those logs alone would constitute some 1,500 pages of printed information. An ordinary search could isolate all the 404 “Page Not Found” errors throughout those pages.
But, Azam argued, just having the event count or distribution or variegation of 404 events would not yield educational information as to their possible cause or remedy. For this, you need machine learning — algorithms that study the traffic patterns on your network when things are progressing smoothly and when they’re not, and are capable of telling the difference.
Events like 404s that are widely distributed over a given interval of time, said Azam, may not be all that intriguing because their causes may be obvious, and their remedies more simple to apply. What’s more intriguing are the negative events that only happen once or twice — events that may be signals of malicious activity, or an anomaly deeper in the system.
Is Nothing Simple?
So to recap this strange but plausible theory: Microservices represent a radical simplification of software architecture that, as a result, could complicate the task of systems management.
Analytics tackles radically increased complexity in systems by reducing their structures mathematically to simple signals. And those signals could tell you what’s going wrong, where, and when.
It’s a theory to which not everyone here at DockerCon this week subscribed, partly because the advocates of containerization are not so eager to portray the object of their affection as complicating the IT department even further than it already is.
But this week has already proven that further complications are almost inevitable, and any kind of simplification would be most welcome.