A network that Doesn’t Suck for cloud and Big Data: Interop 2012 session teaser

Filed in Big Data, Fabrics by on April 30, 2012 5 Comments

On May 8, 2012, I will be giving a talk at Interop 2012 Las Vegas, titled: Architecting data center networks in the Era of Big Data and Cloud.  It’s a 45 minute session (3:30pm to 4:15pm Room K), and it’s a free to all attendees.  If you’re attending Interop this  year, I certainly hope to see you there.  I promise that you will see all new and original material.  There will not be any borrowed or recycled slides.  And at this point there is not a single product slide — this will be an architecture chat.

I thought it would be helpful to give you a sneak peek at what we’re going to talk about.  We’re going to look at a different approach to data center network design.  Something different from the same old premise of centralized, scale-up, big monstrous chassis switches.

For many years network designers have worked with the premise of a two switch rooted architecture.  A pair of Distribution switches, with some access switches underneath it.  A pair of Core switches, etc.  Our hand has been forced this way by the structure of network protocols such as STP with its foundation of a single Root bridge and secondary root.

With the dual root switch premise we scale the access network by building bigger root switches.  We need chassis with more slots, and more density per slot.  When our root switches have reached maximum practical scale we call that a “Pod” and throw a pair of Core switches on top to network in more “Pods” of similar scale, and so on.

That design worked well in the Web 1.0 era of the late 1990′s, and the era of Virtualization 1.0 circa 2005 .  But frankly, that same old network design sucks for the era of Cloud and Big Data.

Moving forward in this era there will be two primary approaches to data center design.  1) You can do it the same old way with centralized scale-up root switches and big monstrous chassis — that will still work to some extent.  Or, 2) you can eschew the old premise with a design based on a distributed, scale out, Leaf/Spine fabric based approach of efficient fixed switches.  All using proven, industry standard protocols with equipment widely available today (and some new stuff too like network virtualization overlays, and maybe TRILL).

When your hand is no longer forced into a centralized scale up rooted architecture, it’s time to re-think which platforms provide the best overall efficiency.  It’s a fact that fixed switches continue to outpace chassis switches in line-rate density and power efficiency.  Space and power, two precious and finite resources.

In the session we will take a look at a sample fabric that scales to 6144 ports, and how cloud and big data workloads can use this fabric.  The distributed design can scale much larger than this, but for a general audience we’ll stick with a general easy to digest scenario.

The Interop session is only 45 minutes, so I won’t be able to provide the technical depth some of you may be looking for.  We will cover the premise and the basics.  I would need almost two hours to really do this session justice.

And on that note, I’m also thrilled to announce that the one and only Ivan Pepelnjak, the  Godfather of network blogging, has graciously asked me to co-host his upcoming Webinar: CLOS fabrics explained.  In this webinar, Ivan and I will be talking in great detail about a network design that doesn’t suck for cloud and Big Data.

Ivan’s webinar will be a full 3 hours of only the best technical detail and explanations we have come to love and expect from Ivan.  And while I do work for a vendor, you can be sure Ivan is going to keep the content balanced, fair, and vendor neutral.  Yes, admittedly my Interop session will be slightly biased, but  of course it will be!  That’s what you should expect from me :-)

Ivan has also allowed me to give the audience attending the Interop session a discount code for this webinar.  So if you stick around until the end of the session I will provide the 20% discount code on the last slide.  This will be the only time the discount code will be shown.

I look forward to seeing you at Interop 2012 Las Vegas, and again at Ivan’s CLOS fabrics explained webinar!

As usual, I will post all of the Interop slides along with a complete and lengthy narrative the week following Interop, right here on this humble blog.

Cheers,
Brad


Disclaimer: The author is an employee of Dell, Inc. However, the views and opinions expressed by the author do not necessarily represent those of Dell, Inc. The author is not an official media spokesperson for Dell, Inc.

About the Author ()

Brad Hedlund is an Engineering Architect with the CTO office of VMware’s Networking and Security Business Unit (NSBU), focused on network & security virtualization (NSX) and the software-defined data center. Brad’s background in data center networking begins in the mid-1990s with a variety of experience in roles such as IT customer, systems integrator, architecture and technical strategy roles at Cisco and Dell, and speaker at industry conferences. CCIE Emeritus #5530.

Comments (5)

Trackback URL | Comments RSS Feed

  1. Daniel says:

    Good basic post, I’ll try to be on you webinar with Ivan!!

  2. Ilja says:

    What always puzzles me in these very large L2 setups is the underlying assumption that server admins somehow would need the *entire* DC capacity for balancing workloads and hypervisor maintenance. If you had let’s say L2 domain of 200 servers and you could not manage capacity within it – maybe it’s time to look for alternative career path. On the other side, apps that span thousands of hosts typically have a shared nothing architecture and fully automated provisioning so VM move is less relevant. Maybe this type of design is really driven by the idea of vlan as a security/multitenancy boundary?

    Will there be a deck/stream available for the interop presentation?

  3. Calvin Miles says:

    Brad -

    Just wondering if you ever posted the slides. I don’t see them here, but perhaps I am just missing them.

    Also, any chance for those of us who couldn’t attend Interop to still get the 20% off code (it never hurts to ask :-) )

    Thanks!

Leave a Reply

Your email address will not be published. Required fields are marked *