Data Center

Jon.Hudson

Of Chocolate & Peanut Butter: SDN & Fabrics

by Jon.Hudson on ‎05-30-2012 10:09 AM (283 Views)

Pic_01.png

Software Defined Networking & Ethernet Fabrics

If while visiting Botswana you try to sell your pristine Teletubbies collection on eBay do you visit a different Internet? If you are in the Australian outback one day and visit www.reddit.com and then board a (very long) flight back to the Center of the Known Universe and pop out your very slick smartphone after touchdown are you hitting the same webservers as you did on walkabout? Of course you are.

Is it the same experience? Not likely. Now is it Reddit’s fault that when you were trying to express your thoughts on the effects of industrialization on third world countries from the top of Burringurrah that your cell coverage was abysmal? Nope. Does it make Reddit look bad from your shiny little cell phone? Yup.

It’s the same phone. You are using things like HTML, TCP/IP, and perhaps 4G in both locations. So all the software layers on top of the hardware are the same at the two locations. Why then is the experience so different?

Because while Mike Holmes can do a lot more with cardboard and duct tape than I can, ingredients matter. You can abstract complexity but abstracting quality is a wee bit more of a trick. No matter how slick Reddit looks if you are getting to it over a 9600 Baud modem it’s going to not be the same as from 365 Main.

Pic_02.png

 

So what about Software Defined Networks (SDN) on a traditional network? SDN will definitely make it better. The Users who send their traffic down Tunnel #1 will most of the time see a nice clean network abstracted from the chaos of hardware. However, if that network loses a node at #2, humans will have to be deployed. The node may need work. Since Spanning Tree is in effect, all your secondary paths are blocked. So to route around this node, reconvergence will be needed, and more human intervention in many cases. If the node at #3 has overloaded links, even more humans will need to be deployed. LAGs may be added, new links provisioned. However this is a manual process that may involve paging someone, waiting for them to drive into a datacenter all before they even start troubleshooting the issue.  And what about the poor soul who happened to be the last person to make a change before this little incident? Just remember, if the CFO can’t update their Facebook status, you will be updating your LinkedIn status.

How easy is it going to be to present this nimble agile software network if it is running on top of a brittle, static, active/passive spanning tree infected medieval network.  How many humans will it take to respond quickly enough to keep this machine working? Do passive 10G links make you happy? What about passive 40G or 100G links?  How does your YouTube video of your cat playing the keyboard look from the Sahara?

Pic_03.png

Enter the Fabric

From the view of the User this can appear to be the same environment. It’s the same pretty tunnel as before. This is the same virtual network giving you VPN access back to the office so you can enjoy the privilege of checking your email while on holiday.

                Under the hood a different story is afoot. In the Ethernet Fabric, if a link becomes overloaded at #2 all a single human (or robot) need do is plug in some addition cross connects. The intelligent fabric nodes will see the new links, auto form LAGs and even detect the locations of the ports in use as well as the length of the cables.  Next they’ll use that data to build frame level trunks allowing for “almost perfect load balancing”. No humans need log into a node. No chance of a tired left hand typing slower than the right leading to possible catastrophe.  If a node at #3 goes offline, you simply lose one of many possible paths. Sure you lose some capacity, but no human interaction is needed, no blocked paths need their cobwebs swept out for first time packet transportation. Replace that failed node, and the other nodes will tell the new device when it connects what is going on and how to join the fabric. No keyboard quick draw is needed.

                Then with the benefits of TRILL and L2 ECMP if another user like #4 up there wants his own virtual network, their flows can be send down different active/active paths for full utilization and efficient use of the those low latency 10G links.  With an adaptive, intelligent, always on active/active physical layer, SDN has the largest number of possible permutations to provide the most agile and nimble end solution to the Users. Communication works better over Fabrics.

Pic_04.png

Even a simple phone call runs better on top of a Fabric. Just ask Mr. Clos. Voices have been laughing, whining and begging over Switched Fabrics Since 1965.  When was the last time you had a call dropped on a landline? It’s really not till we introduced the instability delivered by cell tower coverage and oversubscription that dropped calls came back into common conversation. Can you hear me now?

Just like Chocolate, SDNs can and will be used with great success without Fabrics. And Fabrics like Peanut Butter are and will be used with great delight without SDN to create sticky network meshes.

However unless you are Nutella obsessed, I think we can all agree that just like Chocolate and Peanut Butter, Fabrics & SDN are better together.

Pic_05.png

Announcements