SDN architectures

Cisco Asks the Killer SDN Question

Cisco's unveiling of its Application-Centric Infrastructure (ACI) strategy and plans for its Insieme Networks "spin-in" has raised some core questions for all those interested in the software-defined networking (SDN) debate: Can, or even should, the theory of pure SDN be put into practice?

I won't go into the details of what Cisco Systems Inc. is doing -- my colleague Dan O'Shea has done an admirable job on that front already. (See Cisco's ACI Gets Physical With SDN.)

What is more interesting is whether Cisco's hardware-centric approach -- even if it is driven by an inevitable protectionist streak -- is the one that will, ultimately, make most sense for network operators. Because the debate really isn't about whether this is a SDN play or not. And it isn't just about whether Cisco is looking to lock customers into its technology, though of course that is a major issue and talking point.

The key debate is whether network decision-makers will weigh up their options and decide they are more happy with what Cisco has to offer compared with the alternatives. And this isn't a straightforward issue: Such decisions will be based on personal experiences, finances, skill sets, perceptions, prejudices, and all the other criteria that come into play when human beings (flawed, complex and often unpredictable as we all are) are involved. Only the other week I heard a senior executive from a major mobile operator say that he didn't care if the next-generation technology he was sourcing for his advanced 4G network was proprietary or not -- he just wants it to work.

So maybe the big question, then, is: Will the majority of network operators of any type (datacenter, wide area network, or both) bet their future on conformance to the emerging SDN specifications, standards, and models that are based on open source software and generic hardware?

Cisco, it seems, is betting that enough of them won't walk away from the IP giant with the sometimes intimidating reputation.

Of course, the Cisco pitch was always going to attract criticism. And given that Cisco has said its proposition will only work to optimum performance levels if its hardware (rather than any third-party gear) is deployed, it would be shocking if there wasn't some sort of outburst from the SDN community.

One of Cisco's main rivals, HP Inc. (NYSE: HPQ), was pretty quick to issue a statement attacking the router giant's strategy. The "Insieme ACI poorly addresses market needs" because it is "incompatible and complex," claims HP. "ACI is incompatible with existing Nexus products, and ACI doesn't allow for inevitable migration or provide customer investment protection... Cisco is limiting customers' access to the benefits promised by SDN by locking them into a proprietary and Cisco-only architecture." It concludes that Cisco is "trying to defy the SDN movement with hardware-defined proprietary infrastructure."

Naturally, HP goes on to explain how its OpenFlow 1.3-enabled switches provide "the benefits promised by SDN now."

Here's an alternative, and more neutral, perspective from David Krozier, a telecom network infrastructure principal analyst at Ovum Ltd. .

    Cisco continues to promote the role of hardware in delivering future high performance networks and took great pains to distance itself from pure software-based overlay virtualized networks (like the Nicira technology VMware acquired, Junipers Contrail, and Alcatel-Lucent's Nuage) in the data center. Ovum notes that while the 9000 Series switches can operate standalone, the features provided by the APIC controller require Cisco hardware. While this may raise the hackles of those who believe future networks should be based on generic hardware platforms, this approach is unlikely to match the performance capabilities of ACI.

If you hear someone say "Better the devil you know" in networking circles in the coming months, those uttering that phrase might just be talking about Cisco's ACI.

— Ray Le Maistre, Editor-in-Chief, Light Reading

<<   <   Page 2 / 2
dwx 11/8/2013 | 1:30:08 PM
Re: This reminds me of IMS and SBCs In the US some do, some don't, and many times the RFIs and RFPs may be coming out of different business units.  

NFV is different than SDN.  Running low-bandwidth, high-touch network services on COTS is going to happen.  It happened with VoIP and its related functions and signaling.  For many customers and services we are seeing servers support enough BW to take the place of firewalls and load balancers.   Cisco is embracing NFV, they announced the ASAv the same time as ACI and it's not dependent on ACI.  

Apart from controlling service chaining, the higher speed aggregation and core networks is where SDN comes into play.   At this point Cisco basically re-invented Juniper's QFabric architecture although using an open standard as the switch to switch tunneling mechanism in VXLAN (which came from VMWare/Ncira NSX software overlays they disparage in their ACI presentations) instead of something wholly proprietary.     

The Insieme piece is really being able to communicate application-level data from the switches to the controller so the controller can make intelligent decisions.  The controller has to know where applications live on the network so it's important the provisioning steps run through it. They are using the IS-IS IGP protocol to communicate topology information, so in the end the switches are still running a distributed IP control plane.   I'm not sure how the ACI controller creates static paths across the fabric, VXLAN runs at Layer3 so maybe static routes?  :)    

 The reality is in the leaf/spine datacenter architecture the whole setup is supposed to be non-blocking with 1 hop between endpoints in a single-tier setup.   There is nowhere to reroute traffic... So really the intelligence is in where applications are provisioned and where they are moved when congestion occurs.   Will be interesting to see how everything plays out.

I will agree with Cisco using a "baremetal" switch with a light control plane doesn't really save any money.   The reality is Arista, Juniper, and Cisco just came out with switches using the same merchant silicon as the "baremetal" switches and they aren't that much more expensive.   Customers don't want unproven control plane running their network.  

Alex_Fduch 11/9/2013 | 12:44:39 PM
Re: This reminds me of IMS and SBCs Argument about "proven control plane" for standard decentralized IGP, MPLS control plane and others looks like point of last resort for the vendors like Cisco.

Old control plane proven to be:

- Complex to manage

- Able to create routing loops

- Hard to troubleshoot

- Always needs workarounds, trick and proprietary "improvements" making multivendor solution impossible and locking clients to only one "right choice".

I'm not even talking that traditional vendors like Cisco failed to create full blown EMS/NMS for all their products with good northbound interface.

Of course SDN is not ideal but it allows to break that vendor's jails created by tradional network suppliers and make clients more free in their choices and less dependend.

And for sure nothing comes for free and the price ISPs shall pay is to improve their own expertise and take network knowedge in their own hands. It is a business case to solve and prove.
mbayramo 11/11/2013 | 2:42:45 AM
Re: This reminds me of IMS and SBCs


- The control plane that you are speaking about 25+ year in development cycle already and it works.  The fact to posted message here proofs that distributed computation in scale of Internet works ok. If you move control plane from distributed to centralized point you network you still doing control plane , you still need to compute SPF etc  So argument is completely void in this context.

-Argument regarding a loop ... you can have a loop in SDN environment as well.

In normal layer 3 design you have many tools to avoid that and if you do have loops

it only because of bad design nothing to do with technology.

- Hard to troubleshoot ? so you are saying troubleshooting VM in the cloud and/or programmable interface in between is much easy ? did you troubleshoot 2000 line buggy phyton script ?  

There are many valide arguments around SDN architecture but definitly not those that you listed.


dapperdave 11/12/2013 | 3:40:19 PM
Re: This reminds me of IMS and SBCs Insieme is a product of MPL - Mario, Prem and Luca. All are Cisco veterans. They were the principals at Crescendo (Catalyst switches), Nuova (Nexus switches) and others (MDS storage switching, UCS servers. All are exceptional engineers. And all understand that the amazing engine that makes Cisco successful - the sales team. Nobody sells networking boxes better than the Cisco sales team. Crescendo, Nuova and now Insieme provided the sam\les team with a box-based solution... meaning that hardware is a vessel that carries software. Products are essentially priced based upon the box, not the software in the box. Also consitent with the MPL experience is that they deliver better ASICs than their competitors... this carries the assumption that ASICs are required these days for forwarding performance, deep-packet inspection and other compute-intense packet operations. Arista (CEO is former colleague of MPL at Cisco) is betting against ASICs.

In either case, they are both box-based solutions... dependent upon a sales force that knows how to sell boxes. Because Cisco has a broader range of boxes (controller, switch, compute, security, etc), their sales force will have a better chance selling to enterprise and SPs - because of the "single throat to choke" axiom.

<<   <   Page 2 / 2
Sign In