'We wanted to be Japan's AWS'

Here at ZDNet, we've (hopefully) left few things unsaid about how cloud computing and virtualization have transformed business -- and consumer! -- technology. The terms are deceptively simple, but the implications are far-ranging and complex. There is still much we don't know about how these technologies will ultimately reshape the IT landscape. What we do know: they will permanently change things.

Multinational startup Midokura -- that's Japanese for "green cloud" -- is one of the companies focused on the potential of the enterprise network. It has been around for two years, but until now focused largely on the Japanese market. (It has offices in San Francisco, Tokyo and Barcelona.) The original goal? Be Japan's public cloud, like Amazon is in North America.

Last month, it decided to formally introduce itself and its MidoNet platform to the U.S. market. Midokura hopes that its fully distributed offering will attract companies seeking a more nimble solution -- one that has already been established in another high-tech market.

I spoke with the company's chief strategy officer, Ben Cherian, to find out what makes the company tick.

ZD: This space is getting crowded. What makes you guys different?

BC: There are some market drivers that are causing this product to be necessary. A lot of people are building public and private infrastructure in the cloud. People seem to really be jumping into the idea of competing with Amazon. Control, flexibility, cost-savings -- this is the reason they're building these clouds.

It impacts three different areas: culture, process, technology. Existing networks are difficult to scale, they're inherently insecure and they're [not] resilient to failure.

Three main points about MidoNet. First: we're fully distributed; we believe that is the proper way to handle cloud networking. Second, we believe in the idea of creating virtual networks. Lastly, it's built specifically for the Infrastructure-as-a-Service market, whether public or private.

In order to do a large-scale networking project like AWS, we needed ways that storage and networking would scale. We recruited a team from all over the place -- Dan [Mihai Dumitriu, co-founder and CTO] brought over a lot of people from his Amazon days.

We've been doing this for over two years now, raised $5.5 million from Japanese investors. We have 24 employees across the U.S., Japan and Spain.

ZD: Let's talk about MidoNet a bit. How does it work?

BC: If you look at the bottom-most layer of the network, you have switches and routers. Our software sits on x86 boxes that are basically the compute host and edge-gate host to the cloud. We basically take over the network intelligence of the cloud and present to the user the capability of firewalls, load balancers, switches, routers, NAT devices, L2 and L3 isolation, and so forth.

The underlying network -- you're not actually putting in those devices in the physical network. What we're really doing is, if you think of it as a big circle with hosts and x86 boxes all the way around, the middle of the circle is the network -- we push the network intelligence to the edge of the circle. It allows us to build a network that's much cheaper because we don't have these point machines in between, which don't have good scaling properties.

We don't want these intermediary devices in between. We want to provide the level of services you need as you grow. You can build a very robust physical network with providers at one-tenth the cost of Cisco.

Push the network intelligence to the edge, and build a network that's cheaper and has lower OpEx costs. We believe our model is better because we don't actually touch the physical devices; we're an overlay layer. We require much less of the physical world. All we require is IP connectivity between MidoNet hosts.

Our model is that every host that's in the network running our software is a line card in a global grid router. We feed every machine like it's part of one big router -- if any of those machines fail, we notice instantly and change the topology of the system immediately. We don't send traffic to that machine anymore. We think that's a better model of handling networking in the cloud.

We believe that, inherently, networking was broken because it introduces these intermediary devices that don't scale out. They can scale up, but it introduces service disruptions. But it doesn't scale out. We think it makes more sense to have the knowledge at the edge and treat the physical network as dumb pipes that only need to have IT connectivity.

The idea of having these intermediary devices is a bottleneck. The network itself is a bottleneck.

Our benefit is that your CapEx can be lower and your OpEx can be lower, since we integrate tightly with OpenStack. For the end user, it's seamless.

ZD: Why are you "entering" the U.S. market now, when you've already been here for awhile?

BC: We're entering the U.S. market because there is a market perception that we're a Japanese company. It's inaccurate -- we are there and we started there, but we've had a location in the U.S. for awhile now. Our executives are in both the U.S. and Japan. Because of the roots of one of our founders, Kato-san [As in Tatsuya Kato, co-founder and CEO. --Ed.], the Japanese market understands us very well. We're making it very clear to people that we are in the U.S. We've been here working for awhile.

We started off with the idea that there is a service provider market; they're buying now for financial reasons. There is an enterprise market; they're really just kicking the tires. The people that are talking to us are service providers. In Japan, we're talking to all of them. We've had more and more conversations with service providers in the United States. We think the market itself is going to be geared toward service providers. And they have locations everywhere. From our point of view, we've just been hiring people in the best areas. We're running into deals that we wouldn't be exposed to normally just because of proximity.

We wanted to be the Japanese AWS. Dan moved to Japan to start the company. Kato-san's background was in Japan. And we had datacenter provider partners lined up. It made a lot of sense for us.

We flipped that switch around August 2010 when we shelved our storage product [MidoStore]. We were building technology that was not available; it made more sense to be a technology provider. It fit better within the DNA of the company, too.

The Spain office started because, as we started building out our team, one of Dan's old classmates who he convinced to come onboard had just moved to Spain. So that's how that office got started -- it's primarily a development office.

The U.S. office is partnerships and sales and development.

At the end of the day, we're not known in the U.S. Our goal is to get a wider presence in the U.S. The technology that we have is extremely cutting-edge. The people who are willing to adopt that kind of technology have been the U.S. and Japan. Europe is more conservative. The people that are coming to us are service providers, some enterprise, who have already built a cloud before. They know the problems and they're looking to build their second-generation cloud. There are only a couple of products that really fit the bill for IaaS. But we haven't really won the mindshare war yet.

ZD: Do you feel mindshare is important in this space?

BC: From our point of view, there has to be a lot of market education. As people are learning this, we're running into a much more educated user. We're trying to tell people about the differences of our very technical product. The cloud architect can generally see the benefit of it. The push is to get our message wide so they understand what we're doing.

We're working our sales and product. We're in limited beta right now, but in a few months we'll release our general availability version and we'll announce some customers. Over the next few months, we'll have other partnership announcements -- integration with other stacks and companies. OpenStack is just the first.

We believe cloud networking is inherently broken. As we integrate with these stacks out there, we give users of these stacks the ability to choose a different approach.

Images: Midokura