One size doesn’t fit all

Testers explain that, because of circumstances, varying ingredients were compared.

When comparing multiple products, we as testers generally strive to avoid the “apples-and-oranges comparison” problem by testing similar products in the exact same way. We attempted to do that here, too, but there were several instances in which that just wasn’t possible, either because of vendor demands or device design.

Juniper supplied a system comprising four DX 3600 appliances linked using its ActiveN high-availability system. Redline Networks (acquired by Juniper after testing began) insisted on this configuration as a condition for participating. We agreed, but only with the further condition that we test a single Juniper system as well for an apples-and-apples comparison with other devices. We report both sets of numbers here.

Network managers may like the additional availability and performance facilitated by Juniper’s four systems, but they also should consider the extra cost involved in such a setup. Of course, other test participants with high-availability features probably could have boosted their performance by supplying multiple devices as a single system.

Device design also required some differences in test-bed topology. The Array, Citrix and Juniper devices all required the use of a Layer-2 switch (an Extreme Networks Summit 7i) to move traffic within client and server virtual LANs (VLAN). In contrast, the Crescendo, F5 and Foundry products did not require an external switch, although F5 ultimately requested that we use a switch. And for Crescendo’s device, each port resides on its own IP subnet, so traffic is routed rather than switched between interfaces.

We do not believe the small differences in latency involved (on the order of dozens of microseconds) had any significant impact on test results, but all the same, network architects should be aware that different devices require slightly different network designs.

Scalability test results | Next summary: How we did it >