HP’s ProCurve 2800 series

HP’s latest entries into the stand-alone Gigabit Ethernet switch space – the ProCurve 2824 and the 2848 models – offer a $100 per gigabit port price, but the trade-off is less than ideal performance levels.

HP’s latest entries into the stand-alone Gigabit Ethernet switch space – the ProCurve 2824 and the 2848 models – offer a $100 per gigabit port price, but the trade-off is less than ideal performance levels.

These models sport 24 and 48 copper 10/100/1000Base-T ports, respectively, as well as a typical set of Layer 2 features and limited Layer 3 routing functionality. Each model is housed in single, rack-mountable units with Ethernet ports accessible from the front panel – which is organized well with each port having a LED showing link speed, activity and duplex status. The front panel also has power, fan, redundant power supply (RPS), and general fault indicators.

How we did it Archive of Network World reviews

Subscribe to the Product Review newsletter

The Layer 2 features of these boxes include 802.1Q virtual LAN (VLAN) support, 802.1p packet prioritization and 802.1ad link aggregation. Our tests found that all the Layer 2 features worked well. In particular, the link aggregation tests proved that the switches can bundle several Ethernet links into a logical link for increased bandwidth and fault tolerance between switches.

The limited Layer 3 forwarding could be helpful in tying multiple VLANs to a routed network via a default IP route. For example, the switch can be configured to route between a maximum of eight VLANs. You can configure a total of 16 routes plus one default route. However, there is no support for any dynamic routing protocols such as Routing Internet Protocol or Open Shortest Path First.

Performance was disappointing for both switches.

To quantify a switch’s throughput, we generally measure the maximum throughput it can handle before it starts dropping packets (see How we did it). The 2848 showed packet loss – albeit a miniscule amount – at very small loads. Therefore we couldn’t measure throughput using our typical format. To fairly characterize throughput performance, we ran a packet-loss test with increasing loads.

The 2848 Layer 3 packet-loss test verified the constant small amount of loss seen in our initial test. The loss began increasing at 70% of the maximum load and it increased to 11.52% loss with 1,518-byte packets at 100% load. As the packet size decreased, the loss also decreased (see chart, right).

The Layer 2 packet loss tests on the ProCurve 2848 showed zero loss for all packet sizes with less than 70% of the maximum load. Above this load threshold, packet loss increased until it hit a loss of 11.22% with 1,518-byte packets at 100% load. Again, the packet loss decreased with smaller packet sizes.

The 2824 Layer 3 packet-loss test showed a small amount of loss at maximum load. The packet loss increased with packet size: The largest loss was 0.9% loss with 1,518-byte packets at maximum load. The 2824 Layer 2 packet-loss test revealed similar results.

HP engineers supporting this test said they also experienced packet loss at higher loads, but not to the degree that our tests yielded. They attributed the variation to the fact that we used different gear than they did in our testing.

The latency results – 9 microsec or less for 64-byte packets at high loads – for Layer 2 and Layer 3 tests show a typically acceptable latency level.

Both switches can prioritize traffic by Ethernet 802.1p values. Each port has four output queues with configurable mapping to 802.1p values. The queues are serviced according to a weighted round-robin algorithm. Our tests showed that both ProCurve 2800 models operate somewhat as advertised. Our quality-of-service (QoS) packet-loss test showed that 802.1p values 2 and 3 saw the highest packet loss, and the remaining values were clumped together with a much lower average packet loss. Our QoS latency test showed that 802.1p values 2 and 3 got the highest latency (lowest priority) values 1 and 4 get less latency and values 5 to 8 get a similar amount of low latency (highest priority). Although the results did not show an ideal clumping of values into four distinct groups of packet loss and latency, the 2800 results were acceptable and repeatable.

These switches offer a menu-driven (only used to initially configure the switch) and command-line interface (CLI) available through the console port or a configured management VLAN via telnet for overall management control.

The CLI has a similar look as Cisco’s CLI, although HP has simplified it somewhat. For example, you don’t have to exit the configuration mode to run other commands within the CLI.

HP also offers a Web-based configuration interface. One 2800 switch can act as a master administrative interface for up to 15 other switches. This interface is nicely laid out and offers configuration hints when you’re bringing up the switch for the first time.

You can send configuration files outside the switch via Trivial File Transfer Protocol so you can modify them with a text editor and reload onto a large number of switches. They support a running configuration and a saved configuration that is used when the switch is rebooted. The running configuration is lost when the switch is rebooted.

These switches also support port mirroring to let one port mirror the packets being sent to and from another port. This is a necessity for getting packet traces while troubleshooting enterprise networks.

Reboot takes about 80 seconds, a typical time for this level of switch.

Overall, these switches could be a decent fit at the access layer if you want to aggregate lots of high-speed desktops. Because desktops typically don’t operate at full bandwidth all the time, the performance issues we note might not adversely affect the network.