Storage /

How we did it

By Ken Percy, Network World Global Test Alliance
Network World Fusion, 05/06/02

Performance testing was a collaborative effort with Spirent Communications, which provided equipment and personnel for the duration of the review. We used four SmartBits SMB-6000B chassis, fully populated with FBC-3602A 1-Gbps/2-Gbps Fibre Channel modules. Ports on the SmartBits tool can be configured to send or receive Fibre Channel frames, or both at once for full-duplex data rate testing. Tests were defined and launched using Spirent's SmartFabric Test Application, version 1.10, which then provided port-by-port results.

The participating vendors each submitted one director switch and two 16-port edge switches, which we interconnected in a diamond topology for testing in a two-hop environment. All ports on the edge switches were populated with GBICs of the vendors' choice, as were no more than 64 ports on the director switches. We also tested both vendors' products in a single-switch environment, where the 64 ports were divided evenly between source and destination ports.

For our partial mesh throughput testing, we generated traffic bi-directionally to measure maximum full-duplex throughput. We performed this test a second time, with 16 ports initiating and receiving in a two-hop SAN construct. Using two 16-port edge switches provided by the vendors, we attached half of the ports on each to the director, the other half to the SmartBits targets.

With our high-stress full mesh throughput test, we configured each of 64 SmartBits ports to send large frames to, and receive from, each of the other 63 ports.

We also tested for head-of-line blocking, which analyzes the switch's ability to sustain performance despite congestion on any single port. We split full line-rate traffic from one SmartBits initiator port to each of two ports on the director switches. Half of the traffic, or approximately 50 M byte/sec., targeted one port (Port A) and the rest was bound for the another (Port B). With a second initiator SmartBits port, we sent 100 percent line rate (approximately 100 M byte/sec.) to Port A. This amounted to two data streams of 100 and 50 M byte/sec, bound for Port A, and a third one of 50 M byte/sec. heading for Port B. The idea was to observe how the congestion on Port A effected the uncongested Port B.

With the 15-to-1 throughput test, we sought to observe how each device managed an extreme case of many servers requiring access to the same target storage port. We programmed 15 initiator ports on the SmartBits tool to fire large frames one way at a single destination port, and to record throughput on each initiators. The desired result was even bandwidth distribution along all contending requests.

For our bandwidth distribution testing across inter-switch links we sent large frames from three initiator ports to three target ports on the SmartBits bi-directionally through a two-switch SAN, but with only two trunk links between the switches. The object was to see how the switches distributed three ports worth of traffic through only two trunk links.

In our reboot test, we simply cut off all power to the devices to see how quickly the devices could resume normal operation.

To test component failure resistance, we pulled active modules out of the switch randomly. With only one port module and one CPU module left, both devices functioned normally. Redundant inter-switch links on both devices failed over seamlessly, with full bandwidth restoration when the defeated links were re-inserted.

SAN buyer's guide index