Return to Page 1.

I'm curious what you think is the impact of the webscale companies — the Facebooks, the Googles. Do they end up becoming Cisco's customers, as opposed to the regular enterprises?

Edsall: I think the short answer is yes. We have always looked to certain industries as sort of our leading customers, the ones that really drive our technologies that ultimately benefit all the other customers. Historically, it has been the financial sector. In the last — I don't know how many years; six, seven years? It has transitioned to the MSDCs — the massively scalable data centers, our term for these guys.

They challenge us on automation; they challenge us on scale. If I can meet the scale of one of them, it's really easy to go into a financial guy.

What they don't stress us so much on is — a traditional enterprise is going to have a lot more variety of applications. That's changing with cloud, and we have some important cloud customers. Cloud, you can argue, has every kind of application. But they tend to be — well, "containerized" is a loaded word —

They tend to be in their own little spheres.

Edsall: Yeah, they're insulated inside that environment, so you're not really exposed to the nitty-gritty of the application. That part, I don't get from the MSDCs — the Layer 4-7, the customizations, the variety of applications. So we spend a lot of time with both: the financials and the big guys, the MSDC-scale guys.

What do you think about open source getting into hardware?

Edsall: We support that. We're part of the OCP [Open Compute Project], and we're providing switches. I'm a big supporter of the P4 language that's coming out. In my personal life, I love Arduino and Raspberry Pi and those kinds of things.

But networks are serious things. They are complicated, very difficult. You could say that's our fault.

Um ... you could.

Edsall: In a network, you don't think about the individual switches. There's this big amorphous blob that may be growing or shrinking, and parts of it may be up or down — and it can't go out. You can lose a server. If you lose the network, it's a big deal.

The skills to do the integration of the open source hardware with the operating system and all of those layers — not many companies have the skills to do that. So, if we can provide a service to them, which is to have that pre-integrated, I think there's a lot of value in that.

Will you see some of the MSDCs using an open hardware platform? Absolutely. But I don't think a typical enterprise is likely to do that or that they should do that.

Will you see some of the MSDCs use Cisco hardware? Absolutely. We are trying to be in that space.

What do you think about containers? They're fun, right?

Edsall: Everybody's talking about them, but containers have not arrived. I like them too — they're "fun," if you want to use that term.

The challenge I see for the data center is that the number of endpoints could go up by a couple orders of magnitude. That's a pretty steep increase. It really depends on how quickly they're adopted, but if that increase happens faster than the refresh cycle of the data center, it's going to cause some stress. It means that the current infrastructure will have to support that.

I guess it's the same scaling as always, just in one big step.

Edsall: It could happen very, very quickly. So we've been playing in microcontainer stuff. It's kind of easy. And when you can imagine a container spinning up and doing some work and then shutting down in a second? Wow, we've never had that paradigm before, have we? Computers didn't come and go in seconds. We'll have to adjust to that.

What's going to slow them down is exactly that scale and that ephemeral behavior. If I'm having some class of problem — the application is slow — well, the application doesn't exist any more, so how do I debug it? We'll have to provide telemetry and historical information about what happened an hour ago in an area of the network that we didn't even know was a problem an hour ago. How will we monitor everything continuously? That's an area of great interest for me.

Sounds like you think it's not an insurmountable problem.

Edsall: I think that is achievable.

Easily achievable?

Edsall: No. You gotta be good. [Laughs.]

[Some time after this interview, Cisco launched Tetration, an analytics system that does exactly that kind of historical recording.]