Tuesday, January 26, 2010
SolarFlare's Stern Talks Network Adapters
Last week, Irvine-based SolarFlare (www.solarflare.com) made an interesting move, and started to offer its technology direct to end users. The firm--which makes 10 Gigabit Ethernet semiconductors--said it would start selling its products in its own, branded adapters, particularly to financial and Fortune 100 firms. We caught up with Russell Stern, the CEO of the company, to better understand the move.
What's the idea behind expanding into your own branded adapters--it seems that's a pretty big move?
Russell Stern: Most semiconductor companies have to have to do board designs anyway, for reference designs, and to further qualify them, certify them, and get them through agency approvals. You get almost 80 percent, the you decide what you want to do--do you enter this market on your own, give it to a partner to take to market? What happened to us, is we were doing lots of work in the financial services sector. A particular aspect of the financial services are, is there is lots of market data, feed applications, and algorithmic trading, where those guys are on a quest to have the highest message rate and lowest latency jitter, in order to achieve their performance goals. That's why these applications tend to benefit greatly from 10 Gigabit Ethernet. If you notice, if you dig into the market, lots of the 10G switch manufacturers and high end server guys are all looking to attack that market, because the big banks and investment trading houses tend to be a great beachhead for new technology.
We were working with Wall Street to deploy our NICs, along with the technology we call OpenOnload. What that gives our customers, is a great edge in performance. We were originally calling on those guys in order to get them excited, and to provide a reference base to our OEMs, but what we ended up finding was that they didn't care if necessarily came from an OEM, and in some cases specifically wanted to buy directly from us with support from SolarFlare. At theend of last year, we decided we needed to get into the business, because of that demand. We're now in production with several banks in sixty six different trials across banks in London, New York, and Chicago--at trading houses, Fortune 100 financial institutions--because of what our technology is capable of doing.
What about that technology?
Russell Stern: What we did with our technology, is we focused on performance without the pain. The question is how to get performance out of an Ethernet network--or any network--without having the pain of all these niche, high performance networking technologies that it usually entails, whether that is Myrinet, where you need a special protocol, or Infiniband, where you have to rewrite your applications, or iWarp, where you have to rewrite things to talk to new drivers. That's a lot of pain in an engineered network. What a customer wants is to just have much better Ethernet and UDP multicast. That's what customers are using in this environment. What SolarFlare has, is a highly virtualized interface, which lets you extend 2048 NICs. Why do you need that? The first part, is there is a whole suite of virtualization software--VMware, Xen, HyperV, and all of that. What is happening is in the server market, there's an expanding number of cores, processors, processor packages, bus bandwidth is up, and memory bandwidth is up. When you put applications in a virtualized environment with those things, each application wants scalability across the cores in those systems and servers. What we give you is the ability to have every application that is virtualized scale across hundred of NICs. The biggest hangup with virtaulization, is that your performance can go to hell, so you won't want to put thing virtual in our environment. The piece of technology we did--the middleware we cann Openload, is user level networking, our own TCP/UDP stack, loaded into the user space. An application when it runs can make a standard POSIX socket compliance call to the driver, instead of the kernel TCP stack, and the application now has a direct private channel to the NIC. It cuts all the context switching out and offers basically the same performance as most Infiniband and other engineered networks can offer.
Are you still pursuing embedded designs--you were looking at LAN on motherboard last year, as we recall?
Russell Stern: We still are. We're testing right now with three of the four OEM guys, and hoping to make some announcements over the course of this year. We have ac hip that incorporate a 10GbaseT PHY right within the controller, and that chips is now out. It's sampling and we have customers looking at LAN on motherboard. We're also pushing out some new product that will be announced over the next couple of months, including a new version of our technology using PCIe Gen2 ports, dual port NICs, and a whole host of other products based on the same SolarStorm architecture. We continue to focus on the OEMs and that, but we found that this particular market is really a hotbed and beachhead for 10G technology. It's a situation, where we are finding who really wants and needs 10G, and finding early adopters in the financial services sector. We're trying to serve the market ahead of the OEMs, with a product that can do that. Our competitive benchmarks are surpassed by no others.
So there's no issue with competing with customers?
Russell Stern: No, in fact, they've been encouraging. Most OEMs want to take a completed NIC anyway. Some want something different, saying they don't want to pay for your NIC so they can make it cheaper, and we tell them to go ahead and take our reference design and they can make it. If having my own brand ends up creating brand value in the marketplace, or if they want to keep it under their own brand and that's important, they can do that. What we're doing, is seeing a bit of our shift in terms of a brand equity that is developing on Wall Street.
You're still venture backed, we recall?
Russell Stern: Yes. Oak Investment Partners are our largest investor. We continue to put money into the company. We're not break even, but hope to be next year. We'll raise some more money to launch these new businesses, as well as for our new products which we start shipping this year.