 Hi, this is Stu Miniman with wikibond.org, here with SiliconANGLE TV's live continuous coverage of Dell Storage Forum 2012, live from Boston, Massachusetts. Joining me for this segment is a CUBE alum, Nick Iliadas, who's the CTO of Broadcom's networking and infrastructure group. Nick, welcome back. Welcome, thank you. So we're doing a little focus here on networking here at the storage conference because there's a tight combination between the networking and storage. So we just heard Dario Zamarian talk about how merchant silicon is the foundation for lots of switch interoperability getting beyond proprietaryness in the network. And of course, Broadcom is the leader in this space. So, did you get to hear a little bit about Dario said? And what do you think about Dell's moves in the networking space? I think Dell has been very aggressive and they're really putting their best foot forward in the networking space. They're a great customer and partner for Broadcom. We work very closely with them. And I think they're going to be very successful because as Dario said, they're new in terms of networking and they're putting together this fabulous portfolio with Force 10, Sonicwall, they had their existing product, the Power Connect. So it's really putting together a very complete end-to-end portfolio and now tying storage into it. They're playing with iSCSI and the ecologic acquisition they did several years ago really brought storage and networking together in a big way. Yeah, absolutely. So while Dell's relatively new to the networking piece, you know, from an Ethernet standpoint, Dell's had a lot of components there. And can you speak a little bit to kind of Dell's partnership with Broadcom over the years? Certainly. So iSCSI, which is the one I just mentioned, so we were one of the first companies to work with them to get really high-performance iSCSI, our iSCSI offload on our controllers in conjunction with our switches and their ecologic storage was second to none. That was very high-performance. We worked with them to really tune both the controller, the switch, and their storage controller, their storage array, to have the highest performance in the network. Okay, it works seamlessly because of the collaboration that we've got. And of course most people think of the servers when they think of Dell also. Correct. Is the partnership with Broadcom extend there? Yeah, so with this latest release, the Romney cycle, we've launched seven controllers with them back in March and there's three more coming up very soon. So 10 new controllers, both gig and 10 gig, that are now shipping some of them are the default configuration on the Dell servers. Others are NDS cards, either these pluggable cards or MES cards that go inside of the blade system. So complete offering for gig and 10 gig. And with our offload technology, both iSCSI and FCOE are both offloaded in the controller so that frees up CPU cycles. It also gives you much higher performance than what's available from the other competitors. Yeah, I'm wondering if you could speak to, when I think back when iSCSI first rolled out, people just said, give me a standard neck. I don't really need offloads, I don't need to understand storage. And as we've seen, that there actually is more intelligence that's been baking into the network. And when you think about FCOE, it's, well, maybe the storage stack that I've had for the last decade or so should be what customers go to. But Broadcom and Intel and others have been moving into the space. So can you speak as to, where is the intelligence? Who owns kind of that knowledge and how does Broadcom fit into it? So Broadcom, obviously we were not a legacy fiber channel player, but in developing FCOE, it really leveraged on all our strengths that we had with iSCSI because iSCSI is also a block based protocol. And they're both using Ethernet as their transport. So it was a natural fit for us to be able to develop FCOE on top of what we already done for iSCSI. And using the offload technology, we have very high performance, millions of IOPS per controller, which is far beyond what you can get with a non offloaded controller. And the way we look at this is that frees up the CPU to do more higher value applications. Instead of moving data from point A to point B, the CPU is now available to actually run applications with more virtual machines. So offload really buys you more performance in the server, as well as performance in the network. Okay, but don't all the controllers have some sort of offload? I mean, Intel has it. I mean, looks like Q-Logic have it. You have it. What's the differentiation? Our offload is highly tuned, high performance, and it uses a minimal amount of CPU cycles. So in terms of, if you compare us to the competition, we're sometimes 150% or 200% faster than those other devices. So that plus, the networking plus, Dell's equilogic products, which also have Broadcom components in them as well, work and then seamlessly. Yeah, so you've got a couple of touch points in the network. So Dario, one of the things he talked about is starting to see 40 gig for the uplinks. Where are we with the customer adoption of 10 gigabit and how soon is 40 gig and 100 gig? We're seeing 40 gig getting rolled out now in a significant manner between the top of rack and the end of row. So that tier of the network goes from the access, the aggregation is 40 gig enabled. When you start taking two port 10 gig NIT cards and feeding them to a top of rack switch, now you have capacity that you need to get to the next level of network. So 40 gig, the economics of 40 gig is really compelling because it's actually less than four times 10 gig. If you buy the QSFP cables, you're really buying four 10 gig channels but now they're packaged as a single cable. The other thing about the 40 gig ports is you can actually break them out as four 10 gig ports. So you can actually mix and match on a per, if you buy a switch that has 16 40 gig ports, some of those ports can actually be configured as four ports of 10 gig, some of them can be configured as a single port of 40. So it's very flexible and it's deployment capabilities. Okay, so one of the things I didn't get to ask Dario and I was wondering if you might be able to help tease the part is Dell networking tends to focus on rather scale out architecture. They do have kind of the chassis but most of their pieces are scale out and I believe Broadcom has some solutions that get to really terabits of bandwidth through rather than a single chassis but scaling out. Can you walk us through how that works? I really can't talk about specifically what we're doing with Dell because it's things that are down an ounce so it's actually Dell's job to announce their own products but in general in the industry, Broadcom is out championing what we call the data center fabric and the data center fabric allows high bandwidth horizontal traffic so which a lot of these applications like Hadoop really utilize. And virtualization I would say is very similar. Yeah and actually all the cloud data centers have a lot of east-west traffic so you want to have a fabric that's able to provide that high bandwidth cross-sectional traffic. So we're out there deploying, we have really two solutions. We have a packet based solution and we have what's called the Dune Fabric Cell based solution and it's up to our customer to decide that they want to use a packet fabric which is basically ethernet links set up in a CLOS configuration and multi-path using technology such as Trill. Oh I'm sorry CLOS could you just explain? CLOS, I'm sorry CLOS is the name of the guy who invented this fabric. It looks like a butterfly, it's really a mesh so you basically have many to many kind of connections and the industry term for it is called CLOS. Okay that's different than a leaf spine configuration. Yeah because what ends up happening is you have each box talking to all the other boxes so it basically looks like a cross hatch of connections. Like a fabric. Like a fabric yeah and something people call a butterfly because it looks like a butterfly but it's a fabric and it's called CLOS after the inventor and those links are each independent ethernet links and then you load balance them across them and you use typologies, I mean protocol such as Trill or others to create, you know each path is active so none of them are in standby. Yeah, getting rid of spanning tree and we're tracing it. Getting rid of spanning tree, exactly. So now you have a lot of cross sectional bandwidth. That's a packet-based solution. The cell-based solution is we have a Broadcom acquired company called Dune several years ago and this is a cell-based fabric which provides you even higher bandwidth than the packet-based fabric in terms of 10s of terabits. Then our recent announcements that interop for our 200 gig network processor which is two ports of 10 gig can actually build out theoretically a 400 terabit multi-chassis configuration. Now that's if you go to the max and there's a lot of fiber involved in that but the architectures will support up to 400 terabits which is a phenomenal amount of bandwidth. Yeah, great, it's an interesting architectural discussion if you look on kind of the storage side there's the virtualization piece and then if you look at things like Big Data and Hadoop it's a different architecture while networking we seem to have a good match between whether it's your virtual fabric or going to something like Hadoop is kind of scale out architecture those kind of fit together. The other hot networking topic these days is of course software-defined networks or SDN. What's your take on where we are with that? Yeah, so Broadcom was an early pioneer on SDN we actually worked with Stanford in the early days the reference work, the initial SDN open-flow work was done on Broadcom based switches and still today you can actually order through the Stanford website a reference platform that's running what's called Indigo and then you can run open-flow on top of that. So we are in the forefront of defining products for SDN. We are a member of the Open Networking Foundation, ONF and participate in the work to standardize what we call software-defined networking or open-flow but open-flow I think is just one instantiation of what software-defined networking is. I think there's a broader set of capabilities that are being developed by a lot of companies that fall under the umbrella of SDN. So just saying open-flow I think is doing SDN a disservice. I think SDN is a much broader topic and you can really look and slice it in many different ways. Yeah, no, absolutely and as we talked with Dario it doesn't mean that switches go away and physical switches we still have to have physical switches. Fundamentally, is there change to the silicon that you guys make on the switches going to SDN or is it? Yeah, so one of the reasons we're members of ONF is because we want to make sure that as the standard is being defined that the underlying silicon can actually implement it. If you create a standard in the absence of the people that actually have to implement it sometimes you may not do all the optimal things. So by working with ONF we're making sure that we realize things that can be put in silicon and also help drive the silicon in the direction that's optimized for SDN. So it's really a back and forth kind of relationship. I mean the way I look at SDN it's really creating a platform for people to add value on top of the network. Whereas the past network was really turning ports on and off, setting up ACLs. Now you have the ability to program even higher level functionality on top of the network. One of the best cases that I use is power management. The ability to go out and look at your network as a whole see where the traffic is, see where the traffic is and move virtual machines from less utilized service to more utilized, get higher utilization then turn off parts of your network that are not needed anymore and save power. Okay, so Broadcom being a supplier to the switch manufacturer sits in an interesting place in the market. And while I know you love all of your customers equally we've seen a lot of consolidation in the industry recently. The kind of the verticalization with companies like IBM and Dell and HP. Can you give some thought as to where you see kind of the owned by a company like an IBM or HP versus the independent networking guys? Where's this sit in the marketplace? Who's innovating? I think innovation is occurring on all fronts. I think the big vertical and integrated players are creating solutions and then they have the servers, they have the switches, they have the storage. So they can actually create this end-to-end value proposition. What we see from the smaller independent, sorry smaller independent switches, I see a lot of velocity. So these companies are able to bring out some new products quicker and get to market with some unique features and I think that complements what you get from the vertically integrated companies. So you have the agile, creating some new value propositions, maybe niche market type of solutions where you have the larger OEMs that create the end-to-end solutions that are kind of packaged for the end-user beforehand. Okay, so Nick, you're the CTO of this group within Broadcom. You've been attending a lot of these shows. You know, what's exciting you these days in the networking industry? So virtualization of the network is really exciting because I think we're at the infancy of being able to take the network and segment it into subnetworks that are optimized for the virtual machines that are running at the end points and in the hybrid model where the virtualization actually extends over the wide area into the enterprise. So now the enterprise has an extension into the hybrid cloud where you can actually move your workloads out there and move them back and they're secure, they're isolated from the rest of the things that are happening in that cloud data center. I think that's a very exciting area right now. Yeah, so it's interesting you say that. You know, I think back to the 90s when we had, you know, kind of the service providers and, you know, there's this promise that we're going to be able to move things. You know, our CTO, David Floyer is very hesitant about saying, you know, why do you want to move your data? And it's the bandwidth and the speed, you know, there's only so much you can fit over a pipe. So, you know, what's different now? What technologies are going to enable this so that this isn't just a pipe dream? So the WAN technologies, so 10 gigabit is now widely deployed in most service provider networks and you can actually, if you want to pay for it, you can buy a 10 gigabit link or wavelength right to your enterprise. So, excuse me, that gives you the bandwidth. And then the service provider are now providing the SLAs and the serviceability and the reliability for those networks. So you can actually treat that WAN connection as an extension of your local area network or your data center connection and then work with your data center provider, your cloud provider to, you know, either do a platform as a service or a, you know, infrastructure as a service depending on the level that you need. You know, do you want just bare iron that you don't run your own OS and application on or do you want a platform that is able to run your applications on top of? And I think this gives this, the enterprise's flexibility to be able to not have to deploy everything in-house, move some things into the cloud and also give them some disaster recovery because now you have the ability to move things somewhere where, you know, you can be safe from, let's say, an earthquake or a hurricane or something like that. All right. Well, Niko Yadis from Broadcom, thank you for joining us here again on theCUBE. Always appreciate your insight. And this is Stu Miniman with Wikibon, SiliconANGLE TV's continuous coverage from Dell Storage Forum and we'll be right back after this break. Thank you, Stu.