QFabric data center system succeeds in matching 25% of itself
On the heels of an unflattering report claiming performance issues with new products, Juniper this week issued test results claiming that its new data center fabric switches can achieve 25% of their advertised scalability. The tests, conducted by David Newman and Network Test – which conducts independent testing for Network World – showed that QFabric can support 1,536 10G Ethernet ports without taking a performance hit or adding complexity, according to Juniper.
QFabric’s advertised scalability is up to 6,144 10G Ethernet ports, so the Newman tests showed QFabric can achieve one-quarter of its scalability limit. Nonetheless, Juniper says the results are “record breaking” in that 1,536 10G Ethernet ports is the largest count tested.
Newman tested QFabric against three criteria: performance, scale and interoperability. He used Spirent Communications simulators to pump packets through QFabric at a rate that Juniper says represents the most demanding data center and cloud environments. QFabric held up at 15.3Tbps – that’s 1,536x10Gbps – and all ports were managed as one device, Juniper claims.
The test found that QFabric forwarding delay was less than 5 microseconds for all frames sizes up to 512 bytes when offered loads below the throughput rate. Average latency for Layer 3 traffic is 10 microseconds or less for most frame sizes tested, the Network Test report concluded.
Multicast throughput was close to line rate in all tests, regardless of frame length, and average multicast latency never exceeded 4 microseconds.
For interoperability, QFabric interoperated with Cisco’s Nexus 7010 and Cisco Catalyst 6506-E switches when using link aggregation, OSPF equal cost multipath, and BGP protocols, the Network Test report concluded.
The thing is, though – and this is pointed out in colleague Howard Solomon’s blog from Network World Canada – is that QFabric was only tested against itself in performance and scale. It did not go up against any other vendor’s 1,536 10G port implementation, irrespective of whether those implementations were a single device/system or cobbled together from disparate switches. And even then, QFabric succeeded at only 25% of its advertised scale – when’s the 6,144 port test?
Solomon’s blog hits the key point: The only way an IT department can determine if one company’s fabric outperforms another it to test it in their own environment; against other vendors’ implementations.
Maybe that’s what FBR Capital Markets found when it penned that report on performance issues with Juniper’s QFabric, T4000 and the PTX packet/optical transport system pushing revenue opportunities out to 2013, prompting the firm to downgrade Juniper’s stock: it cited carrier lab tests and other “checks” in concluding that the products still had a ways to go in hitting performance and scalability targets.
It’s likely QFabric, et al, was going up against competitive offerings in those customer trials and losing.
More from Cisco Subnet:
30 events that shaped Cisco in 2011
New Cisco cloud computing framework pushes management, collaboration
Cisco data center forecast very cloudy
Cisco reorgs again, folds net mgmt into new cloud group
Chambers: Cisco’s Q1 solid, but access routers, Nexus 7000 lagging
Cisco taps former VMware exec to head security push
Huawei gunning for Cisco in the enterprise
Cisco upgrades WAN optimizers
Gasp! Cisco and HP collaborate on blade switch
Cisco unveils latest Nexus data center switch
Follow all Cisco Subnet bloggers on Twitter.Jim Duffy on TwitterFollow
