Cisco and Brocade have been going back and forth on power efficiency. What's Cisco's focus when it comes to 'green' IT?
Speaking of virtualization, what are your plans for NeoPath?
Ullal: File virtualization for us is more of a service than a product per se, to help customers with NAS migration, etc. However, long term we're also looking at ways to bring in total data center virtualization of which file may be another component, and there are really no specifics on timing on that.
We heard Cisco plans to release the technology for the Catalyst data center switches, why not in storage? Won't it work in the MDS also?
Ullal: Cisco uses Catalyst switches, the 6500 [model], with NeoPath's file virtualization to offer a complete end-to-end file switching service. So, the two can work together, and we offer services based on that for data migration of files.
So what is that product called?
Ullal: It's the NeoPath product.
But we thought NeoPath products have been discontinued.
Ullal: The NeoPath product has been discontinued, but if people want a service, we're making a distinction between a service and a product. We can still offer it -- if somebody wants file virtualization as a service, we can offer it.
So is that where Cisco sees the place of file virtualization going forward? As a migration service?
Ullal: That's today. Going forward, we see a more holistic [approach to] data center virtualization. We've shared that vision, but no dates for products. But in general, when we bring in an acquisition, especially if it's a smaller product and a technology-focused acquisition, generally it takes 18 months to two years to integrate that technology into a broader vision. We've come out of the standalone appliance, which is what NeoPath had. We'll link it into our 6500 blades and eventually we'll build it into a wholly integrated system. That's not uncommon for what we did with firewalls or load balancers or wide area application services, the box, the blade and then the total integration.
Incipient [Inc.]'s block-level storage virtualization software already runs on an MDS blade, and Cisco could snap them up very easily, but hasn't. Why is that?
Ullal: Our philosophy on these applications is not to be the application vendor itself but to enable them across a network fabric, unless they are a very, very network-centric application. You will see us develop network-centric applications that are so intertwined in the network that they belong in the network, whether it's through partnership, acquisition or developing ourselves. The model will always be twofold -- a suite that is highly integrated and embedded into the switch, like VSANs and some of the data at rest, and data migration options that are inherently part of the network; and another suite, that through APIs can work with popular and well-known existing storage applications.
Why keep those two models? Why acquire NeoPath and not Incipient?
Ullal: It depends on if you're activating the control path or the data path. If you're activating the data path and expecting to forward at a gigabit line rate, you have to integrate deeply into network. If it's running at less than gigabit speeds, and you're shunting off, if you will, to the application for occasional use and then back to the switch for connectivity, then it makes sense to do it through an API. We're pretty thoughtful about how that's done, and also about how customers are deploying it.
On another subject, there's been a lot of talk lately about the Fibre Channel over Ethernet standard. Where's the application for that technology? Who would benefit from it?
Ullal: If you look at the market today, there's billions of Fibre Channel and Ethernet nodes installed. I don't think that'll totally change. But moving forward, at the low end you'll see iSCSI, at the high end Fibre Channel, and in the mainstream you'll see enhanced multiprotocol Ethernet. I think it'll be most attractive for customers who are building a new data center because it's harder to reform the infrastructure in existing ones where all the space is taken, all the power's accounted for, and you have to take out something to put something in. It's a huge constraint. It's easier to deal with during the construction of a new data center when you can plan from scratch but still can interoperate with legacy environments.
Aren't customers trying to build fewer rather than more data centers? Isn't the trend more toward consolidation?
Ullal: I think it's very segment specific and depends on how much they've already spent. This is nothing but my opinion because I haven't done a scientific study on this, but it seems certain verticals and the midmarket are still looking to grow more. In larger enterprises the power issue is huge, and that's why consolidation and virtualization are key issues.
So those large enterprises are looking to stick with Fibre Channel?
Ullal: Well, if they don't have the opportunity to create a new data center, they have to look into what already exists and create enhancements. It may be easier to do in a new data center, but you can carve out a way to do it in an existing data center if you rearchitect.
If you already have Ethernet and already have Fibre Channel and there's a protocol that allows you to combine them, why is it an either/or situation?
Ullal: It's more about integration. Today's integration within an iSCSI and Fibre Channel environment is something that needs to go through very expensive and power-hungry gateways. With Fibre Channel over Ethernet, the gateway function between Ethernet and Fibre Channel becomes transparent and the integration easier than today, but you still have to do that integration either way.
This was first published in June 2007