- 15 Non-Certified IT Skills Growing in Demand
- How 19 Tech Titans Target Healthcare
- Twitter Suffering From Growing Pains (and Facebook Comparisons)
- Agile Comes to Data Integration
Network World - Despite interoperability trials and demonstrations involving alternative data center fabric standards, a non-standard fabric technology is said by proponents to be at the front of the pack.
Multichassis Link Aggregation (MLAG) and two percolating fabric standards, TRILL and Shortest Path Bridging (SPB), are designed to create multiple active paths to work around the limitations of Spanning Tree, reduce latency and facilitate more of an "East-West" traffic flow between server racks.
All three technologies are designed to extend or replace Spanning Tree in Ethernet networks for data centers. Spanning Tree allows for one active path, which can induce latency and traffic flows unsuitable for Ethernet-based data center and cloud infrastructures.
ARE WE THERE YET? Figuring out the data center fabric maze
Interoperability trials have been conducted with TRILL - IETF RFC 5556 and its associated documents --and SPB. Indeed, Avaya, Alcatel-Lucent and Huawei say they've conducted successful interoperability trials with SPB, which is defined in IEEE 802.1aq.
MLAG provides node-level redundancy to the link-level redundancy provided by the Link Aggregation Group specification, defined in IEEE 802.1AX-2008. But even though LAG is a standard, MLAG is not; it is vendor-specific and its implementation varies by vendor, and most support various flavors of it in their data center switches.
As a result, two switches in an MLAG group have to be from the same vendor. It can work with most virtual and network switches, and network appliances, including those running LAG - but two switches that are MLAGd have to be from the same vendor.
So is MLAG interoperable?
It might not matter. Users don't build MLAG groups with multilvendor switches, says Doug Gourlay, vice president of marketing at Arista Networks, a proponent of MLAG for Layer 2 fabrics.
"I've never once seen a customer do that in a production environment," Gourlay says. "How many times have you seen a customer say, 'In this aggregation or distribution tier of my network, I want Nortel on the left and Force 10 on the right'? I've never seen it. People don't deploy networks that way."
Networks can be constructed with multivendor interoperability between tiers, Gourlay says. But within tiers, it's usually a single vendor environment.
"If people were already trying to deploy two different devices from two different vendors in the same location, [MLAG] does limit you," he says. "But I've never seen that. The de facto deployment model we have today is people using two boxes, almost always identical, from the same vendor, and frankly almost always the same software version unless they're in the middle of a change control."
And that may be the key: multivendor devices in the same MLAG group requiring a software change or update would require a network outage, whereas devices from the same vendor running the same version of software could undergo an in-service software upgrade if it is supported by the vendor's MLAG implementation.