AI center stage in new Huawei CloudEngine switch

Huawei is betting on its Ascend AI chip to draw the largest enterprises to its latest high-capacity spine switch. The Huawei CloudEngine 16800 embeds AI into the network fabric.

Huawei plans to release a 400 Gigabit Ethernet switch with a unique AI chip designed to bolster the performance of the largest data center networks. The CloudEngine 16800 will compete with U.S. products in China and non-Western markets where Huawei faces less scrutiny over possible security risks.

Huawei introduced the high-capacity switch this week, with plans to launch it in the second quarter of 2019 to address the needs of the most demanding networks, typically found in cloud providers, telcos, large financial institutions and government agencies. The Huawei CloudEngine 16800 is built to be a spine switch in a leaf-spine architecture, holding up to 16 line cards, each with 48 400 GbE ports.

Huawei is the latest -- but not the first -- networking company to unveil 400 GbE switches, which analysts predict will become the norm in large data centers if traffic trends continue. Cisco plans to release 400 GbE switches in the first half of this year, while Arista started releasing products in the final quarter of 2018.

AI's role in the new Huawei CloudEngine

Huawei's switch is different, because it uses a separate chip to process AI instructions for managing the network fabric. However, how much better the chip makes the switch will depend on the related software Huawei includes with the hardware.

"Even though the hardware is done, and it has the right attributes, I think the key -- when it comes to AI capabilities -- will be the functionality that gets added," said Rohit Mehra, an analyst at IDC. "One has to take a longer-term view on the [potential] capabilities of the switch."

Analysts expect the CloudEngine 16800 to use the Huawei-developed Ascend AI chip, which was introduced in October and is scheduled for release in the second quarter. The processor, coupled with Huawei's iLossLess algorithm, will provide "zero packet loss" to lower network latency and increase throughput, according to the vendor.

Huawei's decision to tightly weave AI capabilities into a data center's switching fabric could significantly improve the speed of the switch and help in identifying faults and configurations that hamper performance, said Fred McClimans, an analyst at Futurum Research, based in Kansas City, Mo. Also, Huawei could tap the switch's AI capabilities to automate everyday networking tasks.

"When you think about switch technology, the future is in autonomous operation," McClimans said. "You want to leverage technology as much as you can to keep that switch performing at an optimal state, so your assets are available 100% of the time."

As a competitor against U.S. networking companies, Huawei's impact is felt mostly in Asia -- particularly within the second-largest global economy, China. Other competitive regions include emerging markets, such as Latin America, Africa, the Middle East and Eastern Europe.

"Where Huawei has access to markets, they do very well," McClimans said. "They're a global market share competitor."

Huawei's challenges

In developed Western countries, Huawei faces many challenges. In the U.S., federal prosecutors are investigating the company for alleged intellectual property theft, The New York Times reported this week. In Poland, Huawei recently fired an employee after prosecutors charged him with spying for China.

Google offices in London
France fined Google $57 million for violating Europe's data privacy rules.

Meanwhile, the U.S. government is pressuring allies to shut out Huawei from the wireless networks telecom companies are upgrading to 5G, arguing the company's ties to the Chinese government make it a security risk.

Australia and New Zealand have already banned Huawei equipment from the fifth-generation networks under construction in their respective countries. Germany is considering a similar ban, the Reuters news agency reported.

Dig Deeper on Cloud and data center networking