Software-Defined

New IP: Just in Time, Just in Place Network Services

by Jon.Hudson on ‎03-14-2015 09:39 PM (5,157 Views)

"It was software; in cyberspace. There was no system core; it could not be shutdown.”

- John Connor (Terminator 3: Rise of the Machines)

 

When it comes to advanced networking, we want what we want, how we want it, when we want it: But as humans, we change our minds…a lot. And we can do this because our wetware can grow, change, adapt, even heal.

 

Ok, so we want products that can morph as our needs do.

 

Business runs at the speed of Humans, but doesn’t have their adaptability. Hardware isn’t that lucky, or flexible. And it’s often full of custom or merchant purpose built ASICs. Which basically means someone made a bet 18-36 months earlier about what technology you would need right now…as in this very second.

 

What happens when the wrong bet gets placed? You already know the answer: It’s another 12-18 months before it gets fixed.

 

But what if I want to try out this cool new technology/idea that came out of a Hackathon last weekend? Or even 5 minutes ago? Well, good luck with that. Anytime that network processor gets asked to do something it doesn’t ALREADY know how to do it will get punted to (poorly performing) software. Or as it’s often referred to in the engineering world: the “slow path.”

 

Ok, so we want a normalized hardware layer so anyone and everyone can write the software and test their amazing ideas.

 

GHz & Cores & DPDK oh My

 

Enter Intel.

 

Ever meet an 800-pound Ant? Ok, then you get the idea. Sure you could build one big crazy cool thing, but why not build 80 cool things and then layer some Crazy cool software on top of that?

 

What about performance? Is 80Gb/s out of a single x86 box good enough for you? No? How about 80 of them? (Remember the ants?)

 

We want uncompromised performance. So we must scale small & wide.

 

You still don’t think the performance is there, do you?

 

Do you by chance have a new TV? Does it have icons like this:

 

TV icons.png

 

Now, what if I was to tell you that to really use the services behind those icons you had to go out and buy a separate physical box for EACH ONE.

 

You would laugh at me.

 

If I said “ but, but!! Twitter runs SOOO much better on dedicated hardware,”

 

You would likely just tell me that you don’t care. . . “Because easy is much more important than fast”

 

Ok, so we want Easy > Fast

 

What do I get for my effort of helping software take over the world?

 

First of all you take project plans like this

Days.png 

and turn them into this

Minutes.png

and you can do that ANYWHERE you need from ANYWAY you want. Just in time, just in place.

 

So what does this look like? New IP.

 

And here’s a vision for what New IP might look like for you:

Datacenter.png

The stats:

Those are virtual datacenters. Stretched and stripped across 55 Racks with 42 Servers each. Each server has 6x40G connections and attaches to 6 adjacent servers. The goal is to allow any-to-any communication with the minimum number of hops. This design is based on the 6D Torus/Mesh work being done by Fujitsu. In this model the server IS the top of rack switch. Each hypervisor becomes its own interexchange. All “host” ports turn inside & virtual so all you have is uplinks and transit. Now any application can be deployed anywhere, moved or scaled anywhere. A giant compute resource that can morph into what you want. What/When/Where network services.

 

Here it is, your moment of Zen

 

If….

We want products that can morph as our needs do.

We want a normalized hardware layer so anyone and everyone can write the software and test their amazing ideas.

We want uncompromised performance. So we must scale small & wide.

We want Easy > Fast.

 

Then

 

You want Software, persistent and everywhere. You want software to eat everything.