One of latest companies to get a cash infusion from famed Silicon Valley venture capital firm Andreessen Horowitz is Mesosphere, a startup that is essentially attempting to build an operating system for data centers (read more about that news here). But it begs the question: Do data centers need an OS?
The idea behind Mesos, which is the open source software the startup is commercializing, is that it can sit above all of the hardware and applications in a data center to dynamically provision and manage workloads and infrastructure. The alternative is having silos in a data center for various applications that are manually deployed. Instead, Mesos is intelligent software that finds the best place for workloads to run, monitors how resources are used and looks to consolidate workloads to run a more efficient data center.
+ MORE AT NETWORK WORLD: Behind the Docker Hype and what containers mean for you |TECH DEBATE: What’s the best way to build next-generation data center networks? +
The technology has some impressive backers. The founders of Mesosphere first implemented the software at Twitter, and later Airbnb. Google is credited with developing this idea through software it called Borg. Now, Mesos deploys applications like Hadoop, Cassandra, as well as manages various other workloads. The appeal is that the software can control on-premises or cloud-based workloads, and even both at the same time. It can run on virtual machines or bare-metal servers. It can even control various types of hardware, not just commodity servers, but highly-specialized CPU instances. Mesos recognizes the traits of this blended environment and automatically provisions the resources to where they will most efficiently be used.
Mesosphere isn’t the only company looking to create a management platform for the data center. Many are tackling this problem in a variety of ways.
VMware, for example, has been talking for more than a year now about its software-defined data center strategy. The basis of this is the company’s biggest strength, its compute virtualization software. But with the acquisition of network virtualization company Nicira two years ago, VMware jumped into the software-defined networking arena. Combined, VMware says its compute and network virtualization software provide the basis for an entire data center controlled by software.
Another company looking to take a data center management approach is IO - which sells modular data centers that are meant to be put together like Lego building blocks as customers need more capacity. (Read more about IO here). The data centers are roughly the size of a trailer that would sit on the bed of an 18-wheeler that can be filled with racks of servers. Groups of these modules are controlled by central software, allowing for one pod to be a highly-available, resilient-to-power-outages node while another one sitting right next to it is for lower-availability workloads, for example.
Then there is a company like Mesosphere - it hopes to have a beta of its product ready in the coming months and then plans to generally release this software by the end of the calendar year. It too manages a data center with a software approach that intelligently places workloads where it is most efficient for them to run.
Is all this really needed? 451 Research Group’s Senior Analyst Jay Lyman says for some companies, yes. Businesses that rely on technology for the heart and soul of their operations - like a Twitter, Facebook and Google - need this sort of stuff. These companies have big data centers and need software to control them. But what about the vast majority of the enterprises out there? Do server farms split across a handful of branch offices at your typically enterprise company need a cutting-edge overlay software?
It depends. These software deployments can bring with them huge benefits. But each company has to consider what their specific pain points are, where they want increased agility and then investigate a solution. Last year Lyman published a report titled “The Automation Spectrum,” in which he argues that different organizations will have varying needs of automation for their operations. For example, more rapid application development is a goal of many organizations today. Businesses want to free their software developers to write, test and launch new applications as quickly as possible. (This report costs money but can be found here.)
There are a variety of approaches along the “spectrum” of automation that an IT shop can use to support faster application development, Lyman explains. For savvy development shops that have access to their own resources, there is a do-it-yourself approach. For organizations that want to completely outsource the hardware necessary to support rapid application development, there are public cloud PaaS and IaaS services. For organizations that want to keep that application development work behind their firewall, there are private cloud solutions available. Lyman’s point is that there are many ways to execute a devops strategy - meaning an application development environment where developers and operations processes are more closely aligned.
Data center management is the same way - there are many ways to manage the increased automation of a modern day data center. For some companies, like a Twitter or Google, a complete overlay approach like a Mesos may be the best answer. There are a variety of cloud management platforms in the market from providers like RightScale, Egenera and others that manage public and private cloud resources and allow for workloads to be automatically provisioned to these environments. For other organizations, perhaps embracing the strategies of vendors they currently work with - such as VMware’s software defined strategy, or Microsoft’s hybrid cloud software strategy - is the best way to roll out increased automation of workloads across devops or broader data center resources management.