Features abound in latest and greatest access switch models

While this test's key takeaway may be the big differences in new features, the good news is that, with a very few exceptions, all switches support the same basic L2/L3 functions.

They're all 1U systems with 48 10/100/1000Mbps ports and at least two 10G Ethernet uplinks (except Foundry's FastIron X448, which is 1.5U high). They all offer basic L2 and L3 IPv4 forwarding features, as well as full support for VLANs, 802.3ad link aggregation and L2 and L3 QoS controls. All even re-mark diff-serv codepoints (DSCP), a best practice when classifying traffic for QoS treatment. (It's not a good idea to trust incoming DSCPs.)

It's beyond these basic features where differences start to appear. For example, Foundry's X448 is not stackable, while all the others are (Foundry has other stackable products in its lineup, but supplied the X448 for this project). MAC address capacity ranges from 8,192 for Dell's PowerConnect 6248 to more than 64,000 on HP's ProCurve. And the Alcatel-Lucent and D-Link switches don't yet support the IEEE's 802.1AB link layer discovery protocol (LLDP), a relatively new standard describing how link partners can exchange capabilities information. The other switches all support LLDP.

Power over Ethernet (PoE), often used at the edges of enterprise networks to drive IP phones and WLAN access points, is another differentiator. While every vendor in this test sells PoE-capable switches, only Cisco, Dell, Extreme, Foundry and HP supplied PoE gear for this project. Cisco's Catalyst 3750E is only device tested capable of delivering power to all 48 downlink ports simultaneously. The others require an external power supply to do so. (We didn't measure PoE power consumption; these figures are from the vendors' responses to our features questionnaire.)

IPv6 support also varies widely. For most enterprises, spotty IPv6 support may not matter – at least not today. But there's a strong and growing probability that IPv6 will matter before switches end their depreciation cycles in three to five years. Even for enterprises with no IPv6 now in place, it's still very much worth considering.

Any switch configured in L2 mode can forward IPv6 packets because it doesn't know and doesn't care about L3 headers. When configured as L3 forwarding mode, all switches we tested except HP's ProCurve can move IPv6 packets between subnets (at least in the software version we tested; HP says current 13.x releases do support IPv6 but we didn't verify that).

That's not the end of the IPv6 story, though. It's important to distinguish between forwarding (moving packets between subnets using direct or static routes) and routing (running a routing protocol to learn dynamically where to send packets). The D-Link, Extreme and HP switches do not support the major enterprise IPv6 routing protocols, OSPFv3 and RIPng. And, as noted, there are major differences in switch management methods over IPv6.

As for multicast over IPv6, the Dell and HP switches don't support either version of multicast listener discovery, IPv6's functional equivalent of Internet group management protocol in IPv4. D-Link's DGS-3650 supports MLDv1 but not MLDv2.

< Previous story: For most switches tested, usability comfort zone lies within the CLI | Next story: Alcatel-Lucent wins green bragging rights in switch test >