I recently sat down with Cisco's [System Inc.] Jayshree Ullal, senior vice president for the firm's DataCenter,...
Switching and Security Technology Group, to discuss some of the key issues facing IT today. She spoke about moving toward managing services, not boxes, about InfiniBand and something called the "data center area network," among other provocative ideas.
The conversation began with her listing five key initiatives at Cisco: consolidation, virtualization, security, application delivery and management. She singled out consolidation as the top trend today.
Jayshree Ullal: People aren't building a 100 data centers, they're going from many to fewer. The burden that places on us with everything from power to rack space to cooling, but also the network components, is to pack more into the rack and add more intelligence.
We've done a lot of work to pack in network intelligence, storage intelligence and even application-level intelligence, disaster recovery and serverless backup and virtualization. The key is that we've taken virtualization to a new level by bringing it on a SAN fabric. The goal is to do the same thing on a compute or server fabric -- make it run in a seamless fashion across the network.
The introduction of a compute-area network like InfiniBand gives us more ability to do that. We view InfiniBand as one of the best technologies to give you ultra-low latency and high-performance grid/cluster computing. But more than even InfiniBand – it can be InfiniBand, it can be Ethernet, it can be Fibre Channel -- more than the I/O we're very, very excited about how you can bring that provisioning and orchestration and virtualization together with the right tools. In other words, VFrame, which is the management tool we acquired through our acquisition of TopSpin [Communications Inc.], is as exciting as InfiniBand as a technology. What you can do now is that you can go all the way from network provisioning to security provisioning down to the server provisioning. The data center managers never had one tool that could do all that.
You can take that virtualization domain, which now is separate islands, and marry them together. That's the second trend, beyond consolidation, the actual provisioning of virtualization components, so that your VLANs, your VSANs, your Vcomputes, work together so that you can provision resources.
What's the biggest challenge to accomplish that?
Ullal: The history of data centers has been fairly siloed. The biggest challenge is an architecture, a design framework, that brings their storage, server and networking vendors, and decision makers together in a more holistic architecture upon which they can make decisions. I'm talking about [the data center] staff. Each one tends to throw servers or networks or storage at the problem, rather than look at them as a collection. I think the biggest IT issue is looking at them in the collection. The biggest issue for us is not whether we can develop great products, but can we develop great data center systems that work in tandem with the server and storage vendors.
The last time the data center had that kind of system was during the mainframe era. Do you think this kind of holistic vision can be accomplished without a single player like IBM today?
Ullal: I think it requires a single point of direction, strategy and architecture. I look at this in three tiers. One tier could be 'who's my systems integrator?' -- and that very well could be an IBM, an EDS [Electronic Data Systems Corp.] or Accenture or one of those -- chances are it isn't Cisco.
Then the second tier is 'how do I construct this design?' And this is where I think Cisco, as the network fabric, plays a huge role.
And then the third is 'what do I put next into this fabric?' -- which may be a set of storage or servers or applications. I do think the systems integrator can play a very strategic role in fulfilling that data center vision but not without having the right foundation across.
So 'constructing the fabric' is where you see Cisco playing the biggest role?
For Cisco to play that kind of role, do you need there to be some kind of open standards for interoperability across the board in the data center? How would you assess that after three years in the storage industry, which is not known for interoperability?
Ullal: Most people get confused that we entered the storage market to build a Fibre Channel switch. We certainly built a Fibre Channel switch. Our entry into the storage market was not just about building a Fibre Channel switch. It was about really providing a data center infrastructure for storage, a critical component of which was Fibre Channel switching. You look at what we've done, we've really married IP, Ethernet and Fibre Channel into one platform … We created not just a SAN network, which is pretty well understood, but a DAN network, a data center area network, where you can connect in on optical on one side and storage on the other and connect into your servers and your classical networking.
From a user's point of view, how would you assess how well that's working these days in terms of their ability to choose the technology that they think fits their needs and then put it into their environment with a minimum of integration on their part?
Ullal: Probably the biggest service we've done to the storage industry is to help them understand that it's not a siloed storage industry. It's actually part of a larger network industry. The MDS platform is as much an IP and Ethernet platform as it is a Fibre Channel platform. So it's brought two sets of decision makers together, the classic Cisco catalyst switching kind of decision makers and the SAN decision makers. And that in itself is a huge habit difference, because these two just went off and did their own service. So from a CIO [chief information officer] point of view, that is the first step in a sequence of steps necessary to build a more holistic data center area network.
Think of the server today. A typical server has two Fibre Channels, two Ethernets, a printer connection, a KVM connection, a power connection. What you've got is seven spigots coming out. Each of these spigots requires their own fabric -- seven fabrics -- does that make sense?
It's great for the vendors, but instead can you envision an environment where you have one, maybe two spigots? I think two is a given because the storage more typically runs on Fibre Channel and the servers more typically on Ethernet, but seven to two is a huge reduction. And ideally, you should have one where whether it's your storage, compute or classical networking, you can keep track of that, you can prioritize over one 10 Gbit link.
Whether it's 10 Gigabit Ethernet or InfiniBand, the idea is consolidation, the idea is virtualization. The idea is that once you get there, you can automate your applications. That is sort of the end goal and obviously there is a lot of legacy here that prevents you from getting to the end goal in an existing situation, but if you can create that in a new data center then I think you've not only had a huge savings in capital costs, but the biggest savings is really in operational savings.
So you have a clear goal of either 10 Gigabit Ethernet or InfiniBand for unified fabrics for data centers?
Ullal: Right. Now we don't have to be technologically religious. It could be InfiniBand, it could be Fibre Channel, it could be Ethernet. A lot of that gets defined by what the real need of the customer is. InfiniBand is clearly the best technology for ultra-low latency. And the question is whether a customer requires ultra-low latency or is Ethernet good enough. Fibre Channel is clearly the best technology for reliable end-to-end storage. The question is whether reliable end to end storage is a requirement in every application or is InfiniBand enough? We try not to dictate the choices but understand the requirements and appropriately provide the network for them.
Click here for page two of "Cisco's Jayshree Ullal touts unified virtualization."