Low latency, multicast scale and microbursts - Arista Networks DCS-7508 Test
The cloud is looming. By 2014, Gartner expects 90 percent of personal consumer devices to have integrated cloud services and IDC predicts cloud computing will create 14 million new jobs by 2015.
As the cloud gains momentum, expectations for data center throughput, reliability, and availability increase and network equipment manufacturers are taking the challenge head on.
While the analysts were declaring the age of the cloud had arrived, Spirent was doing its part to make those predictions come true. Last month, David Newman of Network Test evaluated a building block of cloud computing for Network World, using Spirent TestCenter.
The Arista Networks' DCS-7508 data center core switch packs 384 x 10G Ethernet ports into 11 rack units and delivered impressive test results:
- Switched a record-breaking 5.7 billion frames per second
- Delivered multicast traffic to 4,095 groups on all ports, a record for a modular switch
- Ran at wire speed in almost every case
- Buffered up to 83 MB per port under congestion
Verifying major advances in performance, availability, scalability, and security (PASS) is essential to those who are responsible for building and maintaining the cloud. It comes down to knowing what to test and how to test it.
Given the dynamic nature of the cloud and the as-yet-unknown demands of the next killer app, testing data center switches in a full mesh network is imperative. It is the most demanding possible configuration, both for the test platform and the device under test, but it more accurately reflects the ultimate limits of system performance, something that must be known before deployment.
One item of interest is emerging from recent tests and was validated by the 7508 test. The new de facto standard for latency metrics is single-digit microseconds. The cloud demands it and the test platform must have the ability to measure it—on all flows in full mesh.
Also increasing in importance is the ability to contain the effects of microbursts through buffering, which can be difficult to characterize. The claims of various vendors may be based on disparate methods and therefore of little use in comparing platforms. Until an industry-standard for measuring microburst handling emerges, best practice indicates testing across multiple configurations, including a range of frame sizes and oversubscription patterns (2-to-1, 3-to-1, all-to-1) with all ports, for example 256 to 128 in this case, in order to get an accurate understanding of how the device responds in real-world conditions.
Other key elements of validation for cloud computing include availability and system recovery time (in milliseconds rather than in seconds, as delivered by legacy protocols) and routing scalability.
For more information on the individual tests methodologies, configurations, and detailed test results, see the test report from Network Test—Arista 10G switch: Fast and flexible http://bit.ly/H8Ozz8.