On Monday, we reported on an interesting development in a field of endeavor that ReadWriteWeb doesn’t usually cover: the expanding capabilities of cloud servers for handling real-time voice traffic at carrier-class levels. A report was released showing surging levels of service quality for carrier-grade cloud server switches from Alcatel-Lucent, as the company revealed a new 40-gigabit-Ethernet (40GbE) OmniSwitch 6900 switching module “pod.”
It occurred to us that perhaps this story would be… shall we say, enhanced if we showed you results from the right report. Yes, the one we showed you was a Lippis Enterprises test on Alcatel-Lucent OmniSwitch; and yes, it’s making the rounds and it’s dropping jaws. That part’s fine. But as we were talking with ALU yesterday, we realized it wasn’t the 6900 in that report.
So here we have some news on the report we meant to show you in the first place, and you can leave your jaw on the floor where it was with the first report. The difference here is that the OmniSwitch 6900 (unlike the OmniSwitch 10K) is a 1U rack-mount device designed for top-of-rack (ToR) installations; and although it has fewer ports than the 10K, it’s configurable for 40GbE.
Lippis tested the model X40 in this series, which can be configured for up to 64 10GbE ports, as opposed to the X20 whose maximum is 32. The X40 is said to allow for up to 6 40GbE ports, though an ALU product brochure reveals this configuration to be a “roadmap item.” The firm’s tests generate simulated traffic at different packet sizes, sampling the various latency rates along the way.
The firm characterizes the results here as “consistent latency across all packet sizes at full line rate, a comforting result giving the ALU OmniSwitch 6900-X40’s support of converged network/storage infrastructure.” Lippis must have a gift for understatement, because it’s hard not to notice that as the frame rates go up, latency actually declines a bit.
And in the congestion tests (which was the category we pointed out on Monday), the 6900 certainly didn’t perform any worse than the perfect score delivered by the 10X: 100% line rates for all frame sizes at levels 2 and 3; and full 100% forwarding rates for all frame sizes, compared to 80% for all frame sizes for the 10X.
What does this mean, really? As Lippis analyst Scott Bradner writes, a 40GbE switch provides four times the bandwidth as a 10GbE (obviously), but for less than four times the price. And since data centers are making the switch to newer technologies anyway, especially in order to comply with new standards for power reduction, the 10GbE switches (even the relatively new 10X) could be going out of style.
“Could networking become so fast that it will emulate a computer bus so that a data center operates like one giant computer?” asks Bradner.
“The answer is no, but the industry is getting closer. Today’s 10GbE switches produce 400 to 700 ns of latency. By 2014, it’s anticipated that 100GbE switching will reduce latency to nearly 100 ns… More generally, latency is fundamental to assuring a user’s excellent experience. With east-west traffic flows making up as much as 80% of data center traffic, low latency requirements are paramount as every trip a packet makes between servers during the processes of responding to an end user’s query adds delay to the response and reduces the user experience or revenue potential.”
DISCLOSURE: Alcatel-Lucent is a ReadWriteWeb sponsor.