This week Plexxi’s partner CALIENT Technologies announced it has raised a $27 million round of venture financing to continue producing photonic switching with systems that enable dynamic optical layer optimization for data centers. The combination of Plexxi’s programmable network and CALIENT’s 3D optical MEMS technology makes possible a hybrid packet-optical data center that can scale out existing packet networks with an optical circuit switch network. We think it’s great that investors are backing a company bringing photonic switching into the data center. In a recent Forbes article, Peter Cohan describes why it was so important for North Bridge Venture Partners’ Jamie Goldstein to believe in Plexxi’s solution and understand that his “return on investment can be written on one piece of paper.” Start-up SDN vendors need investors like Jaime to make an impact in networking. Here are a few of my reads this week in the Plexxi Pulse – enjoy!
Network Computing’s Tom Hollingsworth asks whether the software defined data center (SDCC) is all marketing hype or if there something real happening. There continues to be a lot of pressure from the blogosphere to get past marketing content and ultimately label solutions. But aside from CTO organizations that have an explicit job to try out new technology, everyone else buys solutions to problems. The definitions, while interesting for discussion, mean less than what something actually does. Whether or not a data center solution is software-defined is less important than if it is useful. There is already a bit of a calming in the definition wars, which probably peaked during ONS. I suspect we will see companies getting back to talking about what their solutions do. This is ultimately what is behind the use case discussions, and even that will be replaced by customer references as this stuff finds its way into production networks.
Colin at Cyberpunkture says that while companies like Nicira are saying network virtualization will deliver SDN, network virtualization is actually not the best take at managing networks where we can finally holistically manage networks with policy and goals separated from the actual devices. The comments on policy are particularly salient to me because they get to the heart of the shift in networking in my opinion. For decades, the networking industry has viewed the world from the inside out – from the perspective of the network. For all the orchestration promises to come to fruition, that has to be flipped. The application is the most important element in the infrastructure, and the network needs to get information about some applications to better collaborate with the rest of the infrastructure (compute, storage, apps). This requires abstraction, but not just a re-imagining of VLANs, ACLs, and other policy constructs. How this abstraction (and supporting data model) gets built will be interesting. And the fact that it runs on a virtualized infrastructure is a supporting detail, not a defining characteristic.
GigaOM’s Jordan Novet covers Embrane’s news this week regarding new capability in its heleos software that permits network functions allocated to a specific application to get transferred with that application as it moves from one environment to another in a data center. It’s great to see the application taking center stage. The networking world has been insular for longer than useful, and seeing folks take the application to the foreground is exciting for all of us.
Irwin Lazar at No Jitter asks: what does SDN have to do with unified communications? He then says that SDN offers the potential to implement dynamic provisioning of networking resources (QoS/CoS, and bandwidth) to support the needs of real-time applications. I thought the most surprising fact within this article is that 17% of workers are wireless only. Of course I type that from the wireless, so I guess I should be more introspective. Irwin is right in talking about the policy-based approaches of the past. We are already seeing demand for orchestration across compute and storage, in addition to the networking portion. Because SDN will work across silos, this will require a different level of abstraction that is more than just a reinvention of existing network policy. That higher-level abstraction should be in application terms and then translated into infrastructure-specific behavior below. It’s a subtle difference, but I think important if the software-defined data center is to become a reality. I wrote some of this abstraction stuff up recently here.
Patrick Hubbard, senior technical product marketing manager at SolarWinds, provides 7 considerations for SDN in Data Center Knowledge this week. This is a pretty good list and very practical. I was slightly surprised to see more treatment of capital expenditure than operating expenditure. SDN might reduce capital expenditure some, but the white box switch movement alluded to in the first item is really more about building the same networks we have today, just with cheaper boxes. The real drive behind SDN is workflow automation. There are things you can do from a controller that you cannot do (or at least not as easily) in a heavily distributed environment with no global view of resources. This automation is going to impact operating expenditure not capital expenditure. And so you would think that things like rate of change, size of operations staff, and density of operator-introduced network issues might be indicators of who is well suited for SDN.
Matt Hammel, contributor to SDNCentral, asks whether network lifecycle management makes sense. Right now, most of the network OS vendors are looking to a set of reference hardware platforms. But if all the hardware platforms look more or less the same (and they must, else they wouldn’t be commodity and get the tremendous cost advantages being talked about), then you really only have the opportunity to swap one platform for another one that is functionally the same but built by a different Taiwanese manufacturer. I guess this is flexibility, but to what end?