Get out of the data Center network Architecture Puzzle Bureau

While vendors are committed to supporting existing or developing industry standards, users want to deploy a single vendor's data center and their main provider's cloud-switching architecture.

Standards such as Multilink transparent Interconnection (TRILL), Shortest Path Bridging protocol (SPB), and multiple chassis link aggregation (MLAG) are supported by many vendors trying to weaken Cisco's power in the data center Exchange arena. Standard support eases the concerns of customers who are trying to prevent "hold-up" by vendors, but it vendors may not necessarily mix and match multi-vendor switches between a data center and multiple data centers.

"In the data center, customers are trying to find the most seamless path for the data center's key needs," said Zeus Kerravala, head of ZK research at the market study firm. ”

At the same time, Cisco will continue to strengthen and promote its Fabricpath solution to the Data Center network architecture Exchange Domain. Cisco claims that the Fabricpath solution is not based on Trill, although it is a trill extension set. The Juniper network will continue to promote the use of the Bo-chip strategy in its Qfabric products to support multiple active paths and one hop arrival in the data center and cloud environment.

Boton said that the current use of the VCs ' optical fiber technology does not include all the functions of the Trill standard. The company says its VDX data center switch is capable of running in VCs mode or in Classic mode, and is more in line with the IEEE 802.x standard. The data tier executed by the VCs uses Trill, but the control layer is not, using the optical fiber shortest Path First (FSPF) protocol. The protocol is an ANSI standard, and all Fibre Channel storage area networks (Fibre Channel Sans) are used as link state routing protocols.

Boton says that when standardized and available, the Backup link state routing protocol can be supported in the VCS. Trill, chairman of the Task Force on Internet Engineering, Donald Eastlake, said the link state protocol in Trill, intermediate system to intermediate system (IS) has been documented since July 2011, Even the University of New Hampshire began testing its interoperability a year ago.

Eastlake said: "RFC 6326 provides trill support for intermediate systems to intermediate systems (IS). A number of achievements have been made. I think there will be a large number of switches from many vendors this year that will support the Trill control layer. ”

Eastlake did not speculate on the reasons for the manufacturer to defer the Trill data and control layer in its fibre switch to fully conform to the standards.

"Unlike using the control layer, there is no inherent problem in using the Trill data tier." Although they cannot interoperate with the standard Trill control layer, they will not never be interoperable. Some vendors will decide to interoperate in their business decisions, but a vendor may find it difficult to interoperate because of business decisions. ”

HP says it supports Trill and its own smart Resilient Architecture (IRF) multi-chassis binding technology for flattened data center networks. HP has now announced that its new 5900-series top-shelf switch will support both standards.

The Arista Network, the extreme network, and rung are now vigorously promoting Mc-lag's fibre architecture as its switch. Mc-lag is an IEEE standard designed to improve elasticity and uptime and reduce latency by creating active/Active network paths for load balancing and redundancy to replace the Ethernet spanning tree protocol.

Because these network architectures are Ethernet based, Extreme says, they should be as easy to deploy as Ethernet.

"Ethernet is open," said Doug Trading, senior marketing director at Extreme Company. Why is Ethernet architecture not open? ”

Avaya supports the IEEE Shortest path Bridging (SPB) specification for Trill backups in its vena network architecture. The shortest path bridging is based on the IEEE 802.1ah Provider Backbone Bridging (IEEE 802.1ah Provider backbone bridging) Mac-in-mac standard. In the field of telecommunications, this standard is recommended as a Layer 2 backup standard for MPLS in metropolitan Ethernet deployments.

Some vendors are betting on both sides by supporting both SPB and Trill. Huawei is also in this camp.

Juniper chose to continue using the qfabric in its products to implement a labelling mechanism that would enable functionality such as a fiber-optic network within the switch. If the switch itself can function like a fiber-optic network, the network architecture configured with the switches will create a more flat topology, Juniper said.

Andy Ingram, global management director at Juniper Data Center, said: "We didn't run a protocol on implementation." We use hardware to transmit data over the export and ingress of the switch's fibre-optic network. ”

Ingram says this has several advantages: it can be managed like a single logical device, without the need to manage the backbone of the topology and every node of the branch. Because there are fewer ASIC in the data path, the latency is also lower. As the components are reduced, the cost is reduced.

"For architecture switch interconnect cards, we do not require them to be intelligent," Ingram said. We should see the gap between 3kW (each port) and 13kW. Since we don't require it to be a switch, we can make it more efficient, make it faster, and make it cheaper. ”

Ingram said the packets would be tagged and split through the interconnection route. This functionality is built into the Bo-Titan and Trident ASIC for qfabric nodes and interconnections.

According to Ingram, Juniper says its qfabric has 100 customers, more than half of which are in production mode for stand-alone QFX3500 node/shelf switches and/or qfabric interconnection. He also said that at least one-third of customers who had never been in contact with Juniper Products had bought the Qfabric architecture directly.

At present, it is found that qfabric is difficult to achieve its 6,144 10G port/One hop billing in customer testing. However, Ingram said that Juniper was working closely with customers through an extended testing period because it was actually a new Datacenter network architecture.

"We plan to produce in the next 60 days, which are for testing," he said. This is important, it is an important part of the customer's business and is a service provider in the cloud domain. Customers will conduct a comprehensive test within a certain period of time. There are customers waiting for the next version of the launch. For them, the new version has some important features. We will make a lot of production in the next 3-4 months. ”

There are also current analysis that many customers are waiting for the qfabric based on the customized Juniper ASIC.

The New York Stock Exchange is a model of Juniper Qfabric, which has been tested by the NYSE since the launch of Juniper Qfabric. However, UBS analyst Nikos Theodosopoulos said in a recent announcement that Andy Bach, senior technology vice president of the New York Stock Exchange, was an advocate for qfabric, and his departure posed some uncertainty to the New York Stock Exchange's Qfabric deployment.

"Qfabric is still hard to measure, but we see 2012 as a test year," Theodosopoulos said in a statement. ”

The New York Stock Exchange rejected our request for interviews with Bach or other Internet officials responsible for the matter. Ingram said: "Qfabric is exactly what I want." ”

"In our customer base, we have not found any reliability, performance or quality problems with our products," said Denise Shiffman, vice president of marketing at Juniper Platform Systems Department. ”

Ingram said the New York Stock Exchange asked to upgrade only its Juniper EX2500 switch to QFX3500. The New York Stock Exchange currently deploys hundreds of EX2500 switches.

Ingram said this month that Juniper will conduct a proof-of-concept with the New York Stock Exchange at its Solar Valley laboratory to "see if they want to further develop the interconnection of the network architecture." ”

The software defines the network and the main visual components OpenFlow can create illusion. OpenFlow and SDN are considered to make the network configuration more programmatic, allowing the Datacenter network architecture to handle large datasets or large data more flexibly.

OpenFlow and SDN supporters say technology can help hybrid cloud computing services provide bandwidth on demand for data center interconnection, and allow global exchange of large datasets collected from global research projects to help scientists work with global partners; Balance the load in multiple active links in the schema network between data centers or data centers.

It may be possible to threaten vendors, OpenFlow and SDN to specialize proprietary networking architectures in niche markets in the data center and cloud, rather than across end-to-end architectures. In the future they may march across the End-to-end architecture and will abandon their march into the field.

"Theoretically, the Agency network should be given greater flexibility and should be allowed to be more sophisticated," said Kerravala of ZK study. Have you ever heard OpenFlow vendors explain to you that they can do things you can't do on a traditional web? ”

Anyway, after three years of hype over the Web architecture, it's time to show some examples.

"2012 will be a year for customers to start deploying," Kerravala said. I would like to see some examples of actual customers, some quantitative advantages. This year is a year from the sidelines to the action, and as we begin to deploy, a batch of examples must emerge. ”

(Responsible editor: The good of the Legacy)