D-Link Systems Inc.'s relatively low-cost Layer 2 switch, with its 24, 10/100M bit/sec Ethernet ports and a single Gigabit Ethernet uplink, is an overall solid access point switch. But our tests demonstrate that it's got a few price/performance trade-offs. While the typical network in which there is a mix of packet sizes and all user links on average are less than fully utilized shouldn't be affected, these trade-offs still need to be taken into consideration before you buy the switch.
The DES-3225G switch - which began shipping with Gigabit Ethernet support in October - is designed to connect users to a backbone in an enterprise network, not to operate in the core of a network. With a price of less than US$100 per port, the architecture, cost and performance reflect the design goals.
The DES-3225G has 22 10/100Base-T ports, a single 1000Base-SX port and two 100BaseFX ports. The 1000Base-SX port and the two 100Base-FX ports are housed in two removable cards. This provides a reasonable number of Fast Ethernet ports with a Gigabit Ethernet connection to the backbone.
The internal architecture of the DES-3225G is a relatively common design, comprising a series of crossbar switches. Four switch chips each service three groups of eight 100M bit/sec Ethernet ports and a single Gigabit Ethernet port.
Each of these four chipsets is connected to a master switch and has a throughput of 2.4G bit/sec.
The DES-3225G fared well in our performance testing, which focused on unicast throughput and latency, virtual LAN performance and operation, multicast/ unicast interaction and quality-of-service (QoS) testing. We found some anomalies in the results, but these problems were the result of the price/performance design trade-offs we mentioned earlier. For example, the switch operates at less than wire speed with 64-byte packets.
Our throughput testing showed that the switch could handle full wire-speed performance for all packet sizes except for 64-byte packets. With those, the switch could only handle 85 percent of wire-speed performance. The less-than-wire-speed performance was disturbing, until you consider that the switch is designed to connect users to the backbone. It is unlikely that all users will at any time fill the switch with 64-byte packets.
Latency was good for the DES-3225G, with a value of 8.35 microsec for 64-byte packets at 50% of full load. This provides a relatively transparent latency, thereby reducing the time it takes for a packet to make its way through a switch.
One of the switch's most interesting features was its multicast support. The DES-3225G can filter multicast traffic on ports that do not need the traffic. The switch watches for Internet Group Management Protocol (IGMP) packets between the users and multicast routers on the network. When a user sends an IGMP join message for a particular multicast group, the switch allows multicast traffic from that group to be switched to the user's port. The result is much less multicast traffic utilizing bandwidth on the user links. This feature works great. Another benefit is that the switch is more likely to drop multicast traffic before unicast traffic during congestion.
The switch's QoS feature generally operated as advertised. However, we found that under extreme loads, the priority queuing features behaved erratically.
The DES-3225G is designed to handle short bursts of congestion with its small queues. Large amounts of congestion create erratic packet prioritization behavior.
The DES-3225G has the ability to put packets into a high- or low-priority queue. This can be done in two ways. First, the DES-3225G can prioritize packets by VLAN priority. The switch looks at the 802.1Q priority bits in the Ethernet header and places the packet in the low-priority queue if the priority value is zero through three, or it places the packet in a high-priority queue if the priority value is four through seven.
The switch can also prioritize packets by port. In this case, the network administrator can set a port to be a high or low priority, and all packets are placed in the corresponding queue regardless of the contents of the packet.
When we created large amounts of congestion, the resulting traffic prioritization did not get handled as expected. It seems that the switch begins throwing away packets regardless of VLAN priority values.
When we eased up on the traffic and measured packet latency, the switch showed distinct prioritization of traffic. To test the VLAN features, we created several VLANs on the DES-3225G. It forwarded packets properly within the VLANs and contained broadcast traffic within each.
We also looked at the switch's ability to handle bad packets. The DES-3225G properly discards packets with cyclic redundancy check errors, alignment errors and packets that are too big or too small.
The DES-3225G scales well when used as designed. The Gigabit Ethernet uplink port performs at wire speed, but an option for a second Gigabit Ethernet port could give the switch more fault tolerance.
The switch also has a good management offering because it can be configured via serial port, Web interface or SNMP utility. The screen-based menu interface seen at the serial console port is a little cumbersome. We would have preferred a simple scrolling command line interface. Also, some commands took too long to execute. The Web interface is convenient, as a built-in Web server provides access to all the configuration elements found on the console port.
The documentation is some of the best we've seen. The configuration functions are explained in an easy-to-understand and succinct manner. In addition, a section is provided to explain the theory behind the features. The switch installed easily, but the console user interface detracted from the ease of tuning the box for the network.
The DES-3225G is a solid, feature-rich switch when used as designed. For the typical network, the DES-3225G's design limitations should not be a problem, but it's up to you to decide if they will adversely affect your network.
Bass is the technical director of Centennial Networking Labs (CNL) at North Carolina State University (www.cnl.ncsu.edu). CNL tests network equipment and network-attached devices for interoperability and performance, and specializes in test automation. He can be reached at email@example.com.