Startup KulCloud develops software-defined networking (SDN) products and solutions with a focus on solving problems encountered at network borders. Founded in 2011, KulCloud was the first Korean company to launch a highly scalable SDN Controller platform. SDxCentral caught up with KulCloud CTO Sueng-Yong Park to talk about service-oriented architectures (SOA) and KulCloud’s SDN solutions for the enterprise/telecommunication markets.
SDxCentral: KulCloud is a relatively new player in SDN. Tell us about KulCloud’s offerings and how the company was started.
Park: Although our company has only recently become known in the SDN community in the United States, KulCloud started in South Korea in 2011 when we were commissioned to connect a national research network to Internet2 ION Service and develop GENI (Global Environment for Network Innovations) applications on it. Our company DNA includes architects and engineers who have worked for a number of top-tier companies including Juniper, Brocade, Cisco, Samsung Networks, and Aricent.
In 2012, we released the OpenMUL SDN platform, which has consistently topped various SDN Controller benchmarks right from its first release. Our flagship product shipping today is PRISM, an SDN routing/switching solution built on top of the MUL platform. PRISM lets centralized logical routing instances control any set of geo-distributed or local data planes. This makes PRISM an extremely flexible framework for vPE and vCPE environments and provides a seamless upgrade path from legacy to SDN environments. We also are developing customized controller solutions/applications for telecommunications and enterprises.
What was your inspiration for starting KulCloud?
Park: Ever since I moved from the industry to the university in 2007, I have been deeply interested in the idea of service-oriented architecture (SOA). I was particularly impressed by Professor Randy Katz’s 2002 presentation that predicted innovations such as the cloud service broker, overlay-based service (VXLAN, etc), path steering, network fabric, and cloud-based infrastructure for mobile handsets. Amazing insights!
Since 2009, my research has been in the areas of innovative connection services. Services in the sense of an SOA that provides the connections between applications, routers, overlays, services, etc. In other words, the connections between whichever heterogeneity and so-called connection-as-a-service (CaaS). So far, I have found nothing better than SDN for delivering these services.
In our experience delivering network services, we found the core enablers for such services to be the physical and logical devices that sit at the network edges or borders. This includes the boundaries where the SDN network meets the legacy network, where LTE network meets traditional L2/L3 WAN, or where the data center network meets traditional L3 domains. These demarcation points require huge flexibility in programmability and complex administrative policy enforcement. KulCloud started with the vision to make these network border devices open, simple, and flexible by employing the core values of SDN.
Tell us about the OpenMUL controller platform. Who are the target users of OpenMUL?
Park: OpenMUL is an open-source SDN Controller platform. We initially started working on it as a research platform in early 2011 to solve various challenges we faced in real-life SDN deployments. The controller is a key piece of technology to tackle the issues we intended to solve, so we have invested heavily in making its performance among the best in terms of flow download, switch, and application handling.
OpenMUL is targeted mainly but not exclusively for the enterprise. In fact, we are preparing for the distributed two-tier controller for geo-distributed applications. The controller can be easily deployed at the network edge/gateways for applications like service chaining or dynamic policy enforcement, enabling seamless connection to legacy networks.
How is OpenMUL different from other alternatives like OpenDaylight or Open Contrail?
Park: Apart from performance, where stock OpenMUL provides around a tenfold performance benefit compared to ODL, our team has gone to great lengths to make different parts of OpenMUL extremely extensible because we feel there is no “one controller fits all” scenario. Every use case needs specialized controller customizations, and it’s easy to add to the core functionality of OpenMUL.
It is equally easy to add new southbound protocols or a new network service without affecting running services for seamless services insertion. We have worked on providing a very extensive OpenFlow support version 1.0-1.4 and 1.5 is planned soon. As a result of this highly robust and flexible framework, OpenMUL users, including ourselves, are able to develop highly customized and quality controller solutions for different customer POCs/use cases in a short time to market.
Does OpenMUL support protocols other than OpenFlow?
Park: OpenMUL will support protocols like Netconf and OVSDB, as these provide a lot of management flexibility in conjunction with OpenFlow.
You mentioned earlier that KulCloud also has an SDN-based routing/switching solution, PRISM. What problems is PRISM designed to address, and how does it address those problems?
Park: We are trying to tackle the issues faced at network borders, and PRISM provides a way to solve them. It lets you run any legacy control plane protocol to peer with neighboring devices while retaining the original “dynamic policy and network control” features that SDN provides. Making legacy control planes work reliably on available SDN hardware like OpenFlow switches has been a tough task, and our team has worked really hard to make the solution work in a real-world commercial telecommunications/enterprise setup.
In a nutshell, PRISM lets you select a set of supported vSwitch or pSwitch switches, choose to run any centralized legacy L2/L3 protocol outwards, and still use SDN to achieve fine-grained network control.
Why is PRISM ideal for the vPE and vCPE use cases? How is it different from approaches such as Brocade’s Vyatta Router or Juniper vMX?
Park: For vPE or vCPE, the data plane (mostly pSwitch) would be at user premises. Service providers usually want vCPE to be cheap and robust, and switching throughput to be high. Those are tough requirements for vRouter or vMX, as they run on the CPU of data plane. To meet the requirements, vRouter or vMX would need more expensive hardware with more powerful CPU, which may not be good for capex-sensitive customers.
PRISM is different because its control plane runs on top of the controller, clearly separated from data plane (pSwitch or vSwitch), so a cheap vCPE or vPE operated by PRISM provides better throughput than those from other solutions. Lower capex is a huge advantage for CPE.
It differs from other vendors because it operates on 1:N model, so the same software can centrally control and scale 1 to N switches elastically. PRISM gives customers the choice of deploying it as an NFV on top of white boxes or hardware switches. Centralized routing instance management and easier service chaining at vPE also make PRISM superior to other approaches.
Can you tell us about any active proof of concepts (POCs), field trials, or production deployments for KulCloud?
Park: We have done a highly successful GiLAN service chaining POC with a Tier 1 telecommunication operator in South Korea. We also successfully developed a POC framework on top of OpenMUL for SDN-based traffic offload for LTE network funded by the South Korean government. We are in talks for deploying PRISM in another nationwide operator in South Korea. Apart from this, KulCloud’s corporate network has been running OpenMUL and PRISM software successfully for the past year. We are looking forward to a couple of production deployments this year.
Thank you for your time today.
Park: It was my pleasure. Thank you for having me.