SoftLayer CTO Duke Skarda isn't a follower when it comes to virtualized cloud services, says dedicated hardware will still play a role.
SoftLayer has become part of IBM, but the Big Blue Smart Cloud hasn't taken over SoftLayer. It's not suddenly grafting Power servers into its x86 server hegemony. On the contrary, SoftLayer is keeping some unique-in-the-industry cloud computing characteristics that made it a valuable acquisition in the first place.
For one thing, SoftLayer CTO Duke Skarda isn't backing off SoftLayer's special emphasis on dedicated servers -- virtualized or unvirtualized hardware dedicated to a single tenant -- as part of cloud computing. Instead of a multi-tenant motel, he's offering single-tenant chalets.
SoftLayer is known for promoting dedicated hardware in its cloud for big data and big database uses. But the firm's leaning in that direction extends far beyond transaction processing and data management. By a three-to-one margin, its existing 22,000 customers tend to use dedicated servers, provisioned and controlled by the customer, which makes SoftLayer a different kind of cloud than Google or Amazon Web Services.
"I don't live in Silicon Valley. I think the clouderati tend to forget all the things that we've learned over the past two decades," he said as a member of a panel at CloudBeat in San Francisco Monday. SoftLayer is based in Dallas. Skarda studied computer science at the University of Texas, El Paso, and is a veteran of Houston's The Planet, one of the early dedicated server hosting providers and cloud creators. The Planet merged with Everyone's Internet in 2006 to form SoftLayer. Skarda emerged as CTO.
Companies used to run each application on its own server, and in cases where the application is a large one with heavy traffic, using a dedicated server still makes sense, said Skarda in an interview after the panel. This stance requires the provider to charge more; the cost of the server's operation can't be defrayed by many users. But the approach may be a more efficient way to operate, if a customer has an application that regularly needs the resource.
Skarda uses an analogy with solid-state disks. They're still viewed as expensive in some quarters and their lifecycle doesn't have as many reads and writes as spinning disks. But they can make a whole system run faster because the CPU and other resources have fewer idle cycles spent waiting on I/Os and other data transfers. If you need fewer servers because you're getting more work out of your existing ones by adding solid state, can't the solid state cost be justified?
"It's the same idea with dedicated hardware in the cloud," he said. "A large service might be using 20 instances on AWS, but it doesn't need 20. It needs three dedicated servers, and the service will be better and more reliable than before, with fewer moving parts," Skarda said.
Three dedicated servers may have less elasticity -- ability to expand or contract -- than 20 virtual machines, but many enterprise customers have reasons for choosing that option, including security, data privacy and performance, he said.
SoftLayer is not currently an OpenStack-based service, although it's recently added an object storage service based on OpenStack's Swift. Skarda knows IBM is a backer of the OpenStack project. "We're in a process of working through what that means," he said. He realizes the different components of OpenStack are at "different levels of maturity," and IBM will be good at helping enterprise customers iron out discrepancies and implement sets of services.
If it can do it for customers, it can probably do it for an internal cloud unit. "There's not one answer yet" on deploying a complete set of OpenStack services, which means it may happen at SoftLayer later this year or "maybe next year."
When it comes to the Great API Debate going on within the OpenStack project, Skarda sides with those who say the project must develop its own APIs as opposed to following and mimicking Amazon Web Services. "Just acquiescing and selecting Amazon's APIs is not the right answer. If you do, you can then only move at the speed of the API provider," he said. OpenStack must have its own set of APIs.
But it's also foolish, he said, to expect to "bifurcate the world" by forcing cloud implementers to choose between the dominant cloud leader and OpenStack. If the project can include an ability to translate between the two sets of APIs and allow OpenStack users to run workloads in Amazon's EC2 if they choose, it would be a valuable option under OpenStack. "If the community wants it, it can build it," he predicted.
Along with its emphasis on dedicated servers, SoftLayer brings an ability to invoke private networks from each cloud server rather than be connected only to the public Internet. Each server contains four network interface cards, two for redundant Internet connections and two for private network connections. That's over-engineering, according to the dictum that cloud infrastructure is bare bones, standard parts.
But SoftLayer placed many of its 13 data centers, hosting 115,000 servers, in wholesale data center space with a variety of carrier connections available nearby. Many of the data centers reside in facilities built by Digital Realty Trust, which targets such locations.
Skarda said that he's having many more conversations with CIOs since he became part of IBM. "I've talked to more CIOs in the last two months than I did in the previous two years," he said. In those conversations, issues such as the availability of private networks and fine granularity of control over the VLAN are frequent topics. Both have been preoccupations of the IT staff for the last 20 years, he said, and the cloud isn't going to make them go away.
Skarda is going to use that sense to push the operation of the SoftLayer cloud forward as part of IBM. So far, he said, he's under no mandates to convert immediately to OpenStack or add rows of IBM Power servers to the data centers. He wants to continue to concentrate on a SoftLayer version of the cloud, one that includes virtualized, multi-tenant servers, but has resources behind dedicated servers.
That, he said, puts him in a position to talk to the enterprise. "When enough customers ask for a particular type of service, we listen," he said. And he'll keep an ear to the ground to detect whether NoSQL or other big data handling services need to be made available -- running on bare metal.
Multicloud Infrastructure & Application ManagementEnterprise cloud adoption has evolved to the point where hybrid public/private cloud designs and use of multiple providers is common. Who among us has mastered provisioning resources in different clouds; allocating the right resources to each application; assigning applications to the "best" cloud provider based on performance or reliability requirements.