VMware networking exec leaves VMware for OpenFlow startup Big Switch Networks: Here’s why

Howie Xu has been at VMware since he co-invented and helped to launch the vSwitch. After leading the development efforts for VMware's networking for much of the past 9 years, Howie joined OpenFlow startup Big Switch Networks. I had the chance to sit down with Howie to discuss his new role as VP of Engineering for Big Switch Networks, why he left VMware, and his take on the future of networking.

 AF: So Howie, would you mind just starting off by giving me a little bit of your background?

HX: I started at VMware in 2002 and at the time, hypervisor virtualization was still in its early days. So we started with virtual networking and found out that, at the time virtual networking was a new concept and most people simply underestimated the huge impact it was going to make. One thing that really stood out to me was when I started seeing customers put tier one workloads on the ESX server. And it was at that point the customers started expressing desire for an enterprise grade virtual switch. From that point on we realized that we were in a very serious business, as the products we developed back then needed to support tier 1 enterprise applications. So my background in the last nine years has been always virtual switch, virtual switch, virtual switch.

More Background on Big Switch Networks: 

• OpenFlow startup Big Switch Networks lands funding
• Interview with Big Switch Networks Co-founder Kyle Forster

 AF: Can you tell us why you chose to make the move to Big Switch Networks?

HX: Well, the short version is, I'm not just a systems guy, I focus both on systems and networking. And if you look at the customers key pain points today, I don't know anyone else who would not make the same decision in my situation. Because virtual networking has evolved a lot, but a lot more innovation is long overdue on the physical switch side of the fence. There are so many customer pain points that need to be addressed. So, I'm very excited to join Big Switch and I believe we have only hit the tip of the iceberg with networking virtualization and next generation of the networking paradigm.

AF:  Big Switch is a very new company specializing in a new area of networking.  How do you suspect that things will play out for Big Switch in the enterprise networking market, and how do you see the networking market evolving in the coming years?

HX: If you look at a macro level, there are so many things going on. Cloud, convergence, merchant silicon, server virtualization, IT as a service, software defined networking etc. If you look at networking customers today, this is a fairly tough crowd in the sense they are very conservative. But, to my surprise, I have observed that while just a few years ago I did not see interest from networking customers in disruptive technologies. But today, many of them actually are embracing newer disruptive technologies. I think this is changing because there are several new concerns that are arising. One is that the server virtualization cloud made server creation and adfministration so dynamic, so easy, and so fast paced. But there wasn’t much in that model for networking. Network admins got to take care of a lot more tickets, often with no budget increase which has been challenging. So I think network teams are looking for a product that can increase their involvement in new private cloud architectures.   Server teams have this very attractive product from VMware that delivers tremendous business value, resulting in that often, the server guys end up running the show. I  think network teams are looking for a product to make them look like a champion in front of their CIO's, very much like VMware gave to the server admins 6 or 7 years ago. Many network administrators are proactively looking for, if not embracing newer technology. That is simply amazing given how not long ago, everyone was content with 20, 30 year old technology.

AF: I agree that the popularity of VMware and private cloud has taken networking out of the limelight in some ways, and I think there's a lot of desire for networking teams to have a product or solution that has as much visibility and interest from business leaders as VMware. How does Big Switch plan to address this need?

HX: For Big Switch, we focus on customers’ pain points and we prioritize products based on the most immediate customer pain points. I believe that there is so much need for innovation on the physical networking side for the very near future. Networking cannot go towards a siloed approach where there is virtual networking on the server side that is a separate island from physical networking. Management of networking needs to be converged and future products have to enable this kind of convergence. I look forward to driving and delivering innovative products from Big Switch in this space.

AF: There are a firestorm of new technologies competing for the data center, Trill, SPB, DCB, FcoE, 802.1QBH/QBG, NVGRE,  VXLAN, OpenFlow, Q-Fabric, VCS, vn-link, vn-tag, vpath, sr-iov etc… With all of these technologies competing for the Data Center, what do you see as the right path for most enterprises for networking virtual machines?

HX: So the right path from my perspective is to give simplicity, give cost reduction, and give more agility to the customer. I think your observation is right, that there’s so much new technology out there. But if there is one thing the VMware experience taught me, you need to sove a tangible customer pain point and not ride on hype. Server virtualization solved a very clear customer pain point on the server side. We need to do the same thing for networking.

AF: For virtualized networking, access layer network policies in the data center have eroded significantly as VMware has gotten more popular. Do you think there’s a need to be more granular with network services on virtual machines and if so, what are things we can do to accomplish this without impeding the ability to dynamically provision virtual machines and have agility?

HX: Even though the access layer will in a way move to the server side, networking as a whole still needs to deal with policies. QoS is an example. It’s not enough to do QoS on the access layer only; we need to do QoS across the board. So how do you get a network management paradigm that can take care of the whole thing, not just virtual or top-of-rack? Not just look at things as siloed, but give you a more comprehensive view. I think this is a key problem that still needs to be solved. Another thing that’s kind of related, how do you make sure that every time you configure policies on the virtual side, it’s consistent across the board, on the physical side as well? This is another customer pain point that needs to be solved. So essentially, comprehensiveness and the consistency between the virtual and physical side, that’s sort of the pain point that needs to be solved while considering the need for business agility in the sense that you can still move around the VM anywhere, anytime to any place. There is still a technology gap there to prevent this from being a reality.

AF: Longer term, I think it would be ideal if applications could tell the network what their requirements are via API, securely and with the ability to include an approval mechanism in the workflow. What are your thoughts on this?HX: Infrastructure technologies include server, networking, and storage. On the server side, we have come a point where we are using policy-based resource management. For storage, we have come a long way already. On networking, we are very, very far away to get to what you just mentioned. I definitely think we need to move more towards policy based management, and there is so much room for innovation in this space.AF: I think if you look at a lot of the new Data center technologies that are emerging, when I look at a lot of them, I see box pushing. It makes sense for me to want to make every single forwarding decision on a box if I’m the guy that sells boxes. And maybe that makes sense anyways, but the in my opinion all these things need to be approached from a technology-neutral standpoint. For example, in the future data center, what is the best way to deliver security? Does it have to be an access-list anymore? Maybe in a different form or a different place. Maybe the best solution is part network and part application and part client software and part a number of various different things, but I think the problem needs to be approached from a technology-neutral standpoint.

HX: Thats exactly the problem on the server side that VMware has been solving the last ten years. If you think about it, ten years ago, it’s the same thing. Whenever you need a workload, just stack up one additional box. If you need an additional server, you stack an additional physical server. In the last few years, the market has come to realize this is not an ideal methodology and started to move down the server virtualization route. On the networking side, it’s the same thing. You cannot just say, “I want an additional service. I have another box.” That’s very much a siloed way of thinking about things. So we just need to solve the same problem on the networking side.

 AF: In terms of hypervisor switching, there are many different competing visions for how the technology will evolve. Some vendors believe every I/O decision should be made on a physical switch, others are moving additional switching capabilities onto the NIC or CNA card, and many feel a hypervisor switch is more than capable of supporting vm networking. And with OpenFlow there is another paradigm that allows flow caching which can offload hardware requirements from the hypervisor while still supporting increasingly granular network policies delivered from a central point. Of these methodologies, what do you think is the best path for most enterprises?

 HX: I think it all comes down to what kind of problems the enterprise is trying to address? Low latency applications, high throughput applications, virtualized environment? What can and cannot be done for your use case? We may have some technology glut, but customer problems are not yet solved. I'm a firm believer that if you address a key customer pain point by solving difficult computer science problems and presenting the simplicity to the end-users, good things will happen.

 AF: Anything else I missed that you would like to comment on?

 HX: I think the next few quarters are going to very exciting for both big switch and the networking ecosystem. I'm very sure that we are going to roll out more products, more features, and deliver compelling new capabilities.