Microsoft Research distinguished scientist Victor Bahl has been spreading the word about Micro Datacenters, also known by the adorable name "cloudlets," as a key concept for optimizing the performance and usefulness of mobile and other networked devices via the cloud. Service providers have embraced this vision most strongly from the start, but it won't be long before enterprise IT pros will likely do the same, Bahl says.
Here's a more in-depth look at the What, Why and When of mDCs:
I notice that a lot of the research you’re involved in includes not just mobility, but the cloud. Are the two inextricably linked going forward?
Yes, absolutely. We expect more and more from our mobile devices but despite major advances in technology, resource poverty continues to limit the type of applications we are able to run on these devices. Constraints such as battery life and bandwidth are fundamental and not simply a temporary limitation of current technology. To break free of such perennial problems, we must build technology that enables mobile users to seamlessly use nearby computing resources, which have a persistent high-quality connection to the cloud. This way, users can get the benefits of cloud computing without incurring the wide area network delays and jitter that are common in today’s Internet. If we do this, crisp interactive response from immersive mobile applications that augment human cognition will be easier to achieve.
You delivered a keynote address at IEEE WCNC earlier this year titled “Cloud 2020: The Emergence of Micro Datacenters for Mobile Computing.” 2020’s not that far off: What are Micro Datacenters and why should enterprise IT pros care about them?
At the core a micro data center, or cloudlet or mDC, is a rack of servers available at thousands of locations around the world, never more than a few milliseconds away from the client devices. The mDC is connected to the “classical” mega data center i.e. the cloud via a low-latency, high-bandwidth Internet connection. The software running in these mDCs supports multi-tenancy, which means different services from different providers can be supported simultaneously. Client devices including smartphones, wearable computers, and other IoT devices can use an mDC both as a computing resource as well as a caching resource. This end-to-end arrangement allows developers to build new applications and enhance the performance for existing applications.
As mDCs become popular, enterprise IT pros will most likely have to deploy them on premises as cloud accelerators for the services their users depend on. As users come to expect and rely on the high performance and new applications enabled by mDCs, IT pros will be expected to monitor and manage these servers 24x7x365.
What sorts of applications do you envision would be enabled with Micro Datacenters that are impossible with today’s mobile architectures?
Well-known services and applications such as Bing search, Office 365, Azure services etc., will all benefit from these cloud accelerators. In addition, any application that requires heavy use of computation (CPU, GPU, memory) and battery will also benefit. For example vision-based applications, video and sensor analytics, speaker recognition, etc. When wireless bandwidth cost is an issue, offloading computation to mDCs will reduce spectrum usage. By offloading computation, battery life on the end devices will improve as they will do less work, assuming that the energy cost of computation is more than the communications cost.
How drastically do you expect our current mobile tools — smartphones, tablets, early smart watches — will change by 2020?
Wearable computers will be the rage. Applications that benefit from real-time data analytics will be pervasive and vision applications that augment human cognitive abilities will be on the rise. Generally, I believe mDCs will open the door to a new world of disaggregated cloud computing, which will improve the performance of new IoT services and current cloud services.
What role would Microsoft play in supporting such Micro Datacenters? Would the company essentially be getting into the Content Delivery Network provider game, or working with such companies?
In Microsoft Research our goal is to invent technologies that help ensure Microsoft’s future. Microsoft product groups make decisions of what to ship heavily based on their customer needs and their relentless pursuit towards making Azure the best cloud platform in the world. While I cannot comment on what Microsoft product groups will do in the future, CDNs, mDCs and other similar technologies are all within the scope of what Microsoft does.
Where does the most engineering and research effort need to go in order for us to realize this improved mobile computing world?
There is a lot to be done - for example, packaging mDCs to make it easy to deploy them in places where there may not be much IT support; physical security to protect the data that resides on mDC servers and disks; programing framework, to make it easy for software developers to deploy services and applications on a global-scale mDC infra-structure; also management and support for geo-distributed analytics.
There is already a lot of research, which documents the virtues of deploying mDCs close to the user. In recent years executives of several large multi-national IT and telco companies have picked up on it and are describing the benefits of mDCs / cloudlets in their talks. There is even a new ETSI standardization effort on mobile edge computing and new conferences (including the Mobile Edge Computing Conference and The First IEEE Symposium on Edge Computing) dedicated to mobile edge computing. So this is happening.
How do you anticipate enterprise IT pros' jobs changing as a result of where mobile computing is headed? Are there things you would advise them to do, skills-wise, so as not to fall behind?
The Internet is wonderful at connecting people and services. But, it is best-effort i.e. it is not fast, nor does it provide predictable performance. We are becoming more mobile, are doing more, and are context switching more. IT customers are demanding faster access, faster response time, and immediate analysis of large quantities of data which requires more computation and more storage than ever before. Unfortunately, because of the way our network protocols are designed even a few 10s of milliseconds delay can kill user satisfaction. Unless we “fix” the Internet, of which cellular networking is a big part, the only way I see us solving this problem is by deploying thousands of mDCs or cloudlets globally.
In general management of mDCs will be challenging as company executives expect their IT pros to think though the issues and provide smooth accessibility and operation across the different geographic regions their company operates in, even in places that do not have convenient access to the servers. Also, as mentioned previously, IT pros will have to think about how these servers are physically secured. Getting ahead by understanding what mDCs are capable of, their role in emerging cloud and IoT space; asking the right questions and providing honest-to-goodness requirements to mDC providers will lead to competitive advantage. I expect most IT professionals will embrace cloudlets as part of their portfolio of enterprise services they support. The good news is, many engineers understand the outstanding issues and are working on them tirelessly.