SDDC 2014 – Software Defined Data Center
Stephen Foskett opens the symposium by welcoming this he guests and crew. It’s the 3rd Software Defined Data Center symposium and the number of people attending is quite good. The goal of this symposium is to find out what SDDC is in reality. The day schedule looks good with various companies sharing their thoughts.
Michael Dvorkin, Cisco
Passion is to minimize interaction with hardware in data centers. Micromanagement doesn’t sound good when done on people, but in a DC it’s how things work: everything needs to be defined and monitored and managed. A DC grows and grows almost in a linear way. Introducing AI seems the way to go to easy the pain of management. And then came clouds. It doesn’t add much new technology other than giving you access to your VM.
So what is the original intention of this AI, cloud or just about anything new. Let’s not forget the reason of the new building blocks. Defining the application structure is actually about the metadata of that application. By defining the app, you’re defining its life cycle. Each piece of metadata defines how that part interacts with its environment like power, networking services or just about any definable entity in your dc.
The interaction between entities, the interfaces, is what automation is all about.
Intent, capabilities and state, ops constraints, governance.
The granularity of each component defines how precise your app is defined.
Andy Warfield, Coho Data
What the fsck is a SDDC?
What are we trying to define? SDN = separating control from data plane.
Control vs data = desire vs reality
What an administrator desires might not be what he eventually gets or controls. Trying to control it with Software Defined Networking you’re facilitating the matching of the desire with the actual data.
Flash is getting cheaper fast. PCIe $2-2.5 per GB SSD SATA = 0.8 per GB. Disk = 8 ct per GB
1. getting high utilisation out of expensive flash was hard
2. apps do not consume resources evenly, so scaling out was hard to do
Putting flash on the network will slow things down terribly. Or does it? PCIe flash with a queue depth of 32: 235 us (134k IOps). TCP RTT over 10Gb switch: 40 us. So adding flash to the network will provide no problem since the extra latency is only a very small addition to the total latency!
Neela Jacques, Open Daylight
Interoperability is the key to accelerating SDN adoption
Networks are evolving, because it needs business cases, opportunities.
The industry isn’t in agreement yet about how control should be centralised. Do we really need a 100% centralized control or do we need some distributed intelligence like a “special forces” unit in the military? Adding latency to the environment might the wrong way to go, but adding high power distributed intelligence seems a good way to go, but the future has to figure that out or maybe some coexistence will be accomplished.
Innovation leads to incompatibility, but that will lead to even more innovation and eventually well-defined standards. It’ll be a cycle of innovation.
Startup companies provide a way to start over from new, without vendor lock-ins and a fresh way to look at things.