Standard interoperability of Data center architectures

Although interoperability experiments and demonstrations involve alternative data Center architecture standards, proponents of non-standard architecture technology say non-standard technologies are superior.

Multi-chassis Link aggregation (mlag) and two penetration architecture standards (trill and shortest path bridging SPB) are designed to create multiple valid paths to address spanning tree limitations, reduce latency, and facilitate "East-west" flow between server racks.

These three technologies are designed to extend or replace the spanning tree,spanning tree of data Center Ethernet allows for a valid path that can induce latency and transmit traffic that is not suitable for Ethernet data centers and cloud infrastructures.

We have conducted interoperability experiments with TRILL-IETF RFC 5556 and its associated documents, as well as SPB. In fact, http://www.aliyun.com/zixun/aggregation/14423.html ">avaya, Alcatel-Lucent and Huawei also said they had successfully conducted SPB interoperability experiments, SPB were defined in IEEE 802.1aq.

Mlag provides link-level redundancy at the node-level redundancy to the Link Aggregation group specification, defined in 802.1ax-2008, although lag is a standard, mlag is not, it is for vendors, it is deployed by vendors, and most vendors support different mlag in their data center switches.

Therefore, two switches in a Mlag group must be from the same vendor. It works with most virtual and network switches, as well as network devices, including those that run lag, but these two mlag switches must come from the same vendor.

is Mlag interoperability?

Doug Gourlay, vice president of marketing at Arista NX, says this may not be important and users will not use switches from different vendors to build Mlag groups.

"I've never seen a customer do this in a production environment," Gourlay says. "How many times have you heard customers say, ' in this aggregate or distribution layer of my network, I want to use Nortel on the left and use force 10 on the right? ' I've never met, people don't deploy networks like this. ”

Gourlay says that networks can be deployed across tiers across multiple vendor interoperability, but within tiers, typically a single vendor environment.

"If people try to deploy different devices from two different vendors in the same location, Mlag will limit you," he says, "but I've never seen this happen." The deployment model we use now is that people use two chassis from the same vendor and are almost identical to the software version unless they are in the process of change control. ”

This is the key: Multi-vendor devices in the same Mlag group that require software changes or upgrades will need to disrupt the network, while devices running the same version of software from the same vendor can perform an online software upgrade (if supported by the vendor's Mlag deployment).

Shehzad Merchant, senior director of Extreme NX corporate strategy, said that two mlag switches at the same network layer (access/top-level rack, aggregation/distribution, or core) must be from the same vendor because they need to exchange status information.

"Two switches at any given level must come from a single vendor because they exchange status information," Merchant says, "and the Mlag operation requires synchronizing state information between the two switches." ”

Although the way they exchanged information was standardized in IEEE, the Working Group investigated the matter 802.1AXBQ.

But between tiers, switches can come from different vendors, merchant says, because they run standard lag between tiers.

So overall, data center network from server to core to storage can make lag, while Mlag is used for complete active-active redundancy in architecture. Trill and SPB are still developing this functionality and are not finished yet.

"I think Mlag is more interoperable than Trill or SPB," merchant says, "If you look at it from the point of view of the full Datacenter Network, Mlag allows you to interoperate between servers, devices, switches, and storage, which can be achieved by Mlag." Mlag is the only way to help you achieve these. ”

Merchant says that even if the architecture technology is mixed together, if the top-level racks and core switches support SPB or trill, but servers and storage do not support them, you still need to use Mlag to ensure active-active redundancy.

The debate between MLAG/TRILL/SPB and proprietary architectures (such as Cisco's Fabricpath and Juniper qfabric) is only temporary, with consulting firm Ashton, Metzler & Associates Vice President Jim Metzler says suppliers will accept these standards and then add proprietary extensions to differentiate their products.

(Responsible editor: The good of the Legacy)