Inside Advanced Scale Challenges|Thursday, September 21, 2017
  • Subscribe to EnterpriseTech Weekly Updates: Subscribe by email

As Datacenters Bulge, Switch Maker Claims Record 

Cloud migrations, big data and mobility are reshaping the datacenter as more servers are linked and bandwidth capacity joins scaling as essential requirements. Those needs are creating opportunities for network gear vendors as datacenters make the jump to bigger IPv6 pipes.

Among them is Chinese switch maker H3C Technologies, which is claiming a record for switch performance in recent benchmark testing that sought to replicate real-world datacenter operations. Test conductor Spirent Communications recently conducted what it claimed was the highest density 100-gigabit datacenter-switching test in conjunction with H3C and "moderated" by the independent test lab Network Test.

Spirent and H3C claimed the results of the 100G Ethernet networking trial represent nothing less than a "high-water mark" for datacenter networking. Among the findings was stable ("lossless") performance for both IPv4 and IPv6 network traffic at most frame rates using a routing protocol and stressful "fully meshed traffic."

"No task is more important for a datacenter core switch than moving traffic at maximum speed with zero frame loss," also known as dropped packets, the tester noted.

As with most enterprise applications, network latency is a key consideration, often more critical in financial services applications such as high frequency trading than overall network throughput. "While throughput is a measure of switch performance at its maximum speed, latency and jitter [the variation in delaying packets] affect application at every speed, regardless of load," the Spirent test report notes.

H3C's switch (designated S12516X-AF), as with virtually all datacenter core switches, employs a "store-and-forward" approach that caches an entire incoming frame before determining where to forward it.

Spirent said its 100G Ethernet datacenter switch test measured latency, jitter and network throughput simultaneously, using a total of 768 100G Ethernet interfaces and BGP (Border Gateway Protocol) routing.

In tests using frames of at least 102 bytes, the H3C switch was found to move IPv6 traffic at "the theoretical maximum rate" to all 768 100G Ethernet ports without frame loss. Moreover, the tester used "fully meshed" network traffic, which it defined as "the most stressful possible load."

[Chassis with 768 100G interfaces are currently considered the highest performing core switching equipment for enterprise datacenters.]

As datacenter operators attempt to scale their network operations, Spirent stressed: "A key question in migrating to IPv6 in the datacenter is whether core switches can deliver the same latency and jitter as for IPv4 traffic." The benchmark testing found little difference in performance between the two Internet protocols, asserting that there would be little or no cost association with upgrading datacenters to IPv6.

Based on its test results, core switches such as H3C's demonstrate that such a networking capability is now within reach.

Spirent said it oversaw the networking switching test in June with Network Test, an independent testing entity "authorized" by H3C.

H3C is part of a Beijing-based joint venture called New H3C Group, formed last year by Hewlett-Packard Enterprise (NYSE: HPE) and Tsinghua Holdings, the asset manager of Tsinghua University. The Chinese company previously acquired a majority stake in HPE's data networking unit.

About the author: George Leopold

George Leopold has written about science and technology for more than 25 years, focusing on electronics and aerospace technology. He previously served as Executive Editor for Electronic Engineering Times.

Add a Comment

Share This