As data center networks grow more complex, the core needs more than just bigger boxes -- although those come in handy, too. Core switches must also be part of a more resilient, efficient and intelligent architecture that can support highly virtualized data centers, cloud computing and advanced network services.
HP's FlexFabric 12900 series core switch, this month's SearchNetworking Network Innovation Award winner, tackles those challenges with 36 Tbps of switching capacity, a lossless Clos architecture, a path to software defined networking (SDN) and support for a standards-based implementation of the Transparent Interconnection of Lots of Links (TRILL) protocol -- a multipathing alternative to Spanning Tree Protocol. The series comprises two models, the 12910 and the 12916.
SearchNetworking Features Writer Jessica Scarpati spoke with John Gray, global product marketing group manager for FlexFabric, routing and management at HP Networking, about the 12900's capabilities and the benefits of a standards-based approach to TRILL.
The FlexFabric 12900 is a big switch in terms of capacity, but can you talk about some of its other capabilities, specifically how it supports the TRILL protocol?
John Gray: Yes, it's a very large-scale, high-performance, multi-terabit switch that's designed from a speeds-and-feeds perspective to support a lot of capacity and a lot of scale. But what's really unique about this switch is at the time it was launched, it was the first switch to support some open standards protocols, specifically TRILL and SPB [Shortest Path Bridging], and what's unique about this is it's a technology that allows for multi-vendor, open network virtualization. It allows the scale-out of the core in a very seamless way -- certainly with HP gear -- but also because it's open, it allows a customer to have a multi-vendor environment if so they choose.
The thing about TRILL is up until that point [of standardization], a lot of vendors … had their own proprietary way of connecting their switches together in a virtualized way. TRILL and SPB really allow customers to do it in an industry-standard way, which is very important for a lot of customers. They don't want to get locked in any closed environment, and the fact that HP is able to support this in a very open way is a big deal for customers.
If HP is currently the only vendor supporting a standards-based implementation of the SPB and TRILL protocols, then is a multi-vendor core a reality yet? If not, then where is the benefit for customers?
They're very linked, actually -- the fabric and SDN, specifically in the data center. They're almost inextricably linked.
Gray: There are other vendors in the industry that may or may not support open standards. There are other vendors that actually support it but don't sell it that way; they would rather lock customers into a particular scenario. If push comes to shove in a competitive environment, some competitive vendors will either add that support or unveil the fact that they can actually do it but don't recommend it.
We tend to force the issue with [other vendors for] a lot of customers in terms of the way that we progressively support this. What we're dealing with a lot of times is an entrenched competitor that wants to support the proprietary protocols before they'll admit [they can support the standards-based approach] or add the support to their switch or their router, because they know that once they do that, they'll have opened up the door.
We tend to drive the agenda … around open standards support for the benefit of customers so they cannot get locked in, and quite frankly, a lot of customers have been locked in for the past couple of decades. But TRILL, SPB, OpenFlow and BGP are open protocols we continue to push big-time. Being the first with TRILL and SPB, we're forcing the issue with a lot of our customers, and at some point in time, other vendors catch up and they might add that support if they're so inclined.
Is any other type of TRILL- or SPB-based interoperability possible at this point?
Gray: A lot of times we'll see an RFI [request for information] for a particular opportunity come up in the data center, and it won't be necessarily maybe to replace the core. There may be some existing core switches there -- maybe HP, maybe not HP -- but the customer has insisted on standards. There may be an opportunity at the access layer to place a bunch of other switches in the rack or at the top of rack. In that case, it's important for a customer to say … one of the requirements [is a standards-based implementation of] TRILL or some other standards-based protocol so that there is interoperability between layers. … It's between the [switching] tiers that you can mix and match if you have that standards-based interoperability.
In addition to being the first core- and aggregation-switch vendor to support OpenFlow 1.3, what else sets apart the 12900 from HP's competitors?
Gray: Ultimately what SDN is about, from my perspective, is just making networking simpler -- trying to take out the manual process of having to configure hundreds or thousands of lines of code from the command-line prompt. And HP has an SDN strategy that allows customer to be able to ultimately deploy SDN with protocols like OpenFlow and dramatically simplify the operations of the data center with products like the 12900.
In terms of the hardware -- the port density, the speeds and feeds, and the latency -- every six months, new merchant silicon ASICs get developed and all the vendors rush to produce a product to market faster. I think we've caught [the ASICs] market at an appropriate time to trump the competition with this class of product.
At the time of launch, which was around Interop, our main competitor in this regard was Cisco with the Nexus 7000. [The 12900] was a product that offered double what a comparable Nexus 7000-class product could support -- up to two times the switching capacity and up to three times the 40-gig [port] density. I'm comparing it to a Nexus 7010 or 7018, which would be the like-for-like comparison. The speeds and feeds part of it, the scaling part of it, the flexibility part of it -- we continue to trump our competition in that regard. But I think in terms of differentiation moving forward, [it's] the fact that this is baked into a broader strategy with our customers, not just to address their 10- and 40-gig high-scale requirements in supporting tens of thousands of switching ports at scale -- that's important -- but it's [also part of] HP's broader strategy of ultimately trying to remove the human middleware in the data center that customers are struggling with. We're doing that through delivering a software-defined fabric in the data center. This product, the 12900, as well as our other FlexFabric switches and routers, are really at the heart of what I would call a software-defined network fabric.
I always thought of fabrics and SDN as separate spheres. What do you mean by 'software-defined network fabric'?
Gray: Data centers today have dozens, if not hundreds, of different elements that the administrators -- the folks managing that network -- physically have to go off and touch each of those devices on an ongoing basis to configure it, to tweak it, [and do] a lot of manual interventions to actually operate it and deal with it on an ongoing basis. The software-defined fabric I'm describing [uses] a product like the 12900, along with other products that we have, [to automate those functions]. When we deploy a fabric that supports 100 servers, 1,000 servers or 10,000 servers, typically the way these [fabrics] are deployed is with two 12900s in the core and another layer … that connects to the servers with a product like the 5900 [top-of-rack switch series]. That's what makes the fabric. It's an Ethernet fabric.
The software-defined part of that is in … [our] ability to separate the control of that network from the physical forwarding of the network, so all the control is done in software. Instead of having to go out and touch 100 or 200 separate devices every time you make a change … we can all have that done within a separate SDN controller, which is a separate product that we have, and through software we can take away that manual intervention and do it within software automatically in a very programmatic way.
So, they're very linked, actually -- the fabric and SDN, specifically in the data center. They're almost inextricably linked, from HP's perspective. They're not two separate things.
This was first published in December 2013