2 architecture – HP 2424M User Manual

Page 11

Advertising
background image

HP ProCurve Switch 4000M / 8000M / 2424M / 1600M Reviewer’s Guide

• Spanning Tree Protocol (802.1D) support, including “fast” mode to allow support of IPX client

environments (see the application note “HP Switches - Controlling Network Traffic” at

http://www.hp.com/rnd/products/solutions/techlib/techlib.htm

for more details).

• Year 2000 compliant

2.2 Architecture

2.2.1 Store-and-Forward

There are two techniques by which switches move packets from the source to destination ports, 1)

store-and-forward or, 2) cut-through. In store-and-forward, the entire packet is received and stored in

memory by the switch before any decision is made as to the destination port for that packet. For cut-

through, the switch makes the destination decision and forwards the packet as soon as it has the

destination and source address of the packet. Since the destination address is contained at the

beginning of the packet, it is possible to have the packet being sent out the destination port before it is

entirely received on the incoming port. No CRC checking is done on the packet with cut-through

switching.
The HP ProCurve Switches 4000M/8000M/2424M/1600M use a variation of the store-and-forward

method of packet forwarding. The benefits of the store-and-forward approach over cut-through are:

• Allows multi-speed switching: Store-and-forward is the only way a switch can deal with

differing speeds on its various ports.

• Filters out bad packets: Packets that are too short (as well as collision fragments), too long, or

have bad checksums are not forwarded, preventing the proliferation of bad packets.

• Congestion management: All switches deal with packet congestion by using store-and-forward

methods.

• Priority Queuing: QoS priority queues work through a store-and-forward mechanism.

The usual concern with the store-and-forward method compared to cut-through is higher packet

latencies through the switch. The HP ProCurve Switches 4000M/8000M/2424M/1600M have a novel

backplane design that keeps packet latencies among the lowest in the industry for store-and-forward

switches. This is discussed in more detail in the Backplane Design section below.

2.2.2 Backplane Design

The HP ProCurve Switches 4000M/8000M/2424M/1600M has backplanes controlled by an HP-designed

programmable ASIC. Backplane speeds for the various switches are: 4000M/8000M: 3.8 gigabit/sec,

1600M: 3.5 Gbps, 2424M:

3.2 Gbps. The switches support

10,000 MAC addresses. Unlike

many other switches on the

market, data is stored on the

incoming port module and only

travels across the backplane

once on its way to the outgoing

port module. At a measured
latency of only 3.5-12

µsec (LIFO,

100Mbps speed), the

HP ProCurve

ASIC

Controller

3.8 Gbit switch bus

Slot

Module

Slot

Module

Slot

Module

Switches 4000M/8000M/2424M/1600M have low latency for a store-and-forward design. Latency is the

same for both inter and intra-module destinations.

For broadcast and multicast packets, LIFO latency is still at 4-21

µsec

for all ports

, and the variance in

broadcast and multicast latency port-to-port in these switches is

±1.2 µsec, which is very low. For

©1998, 1999, 2000 Hewlett-Packard Co

Revision 3.2b – 1/15/2000

Page 11 of 36

Advertising
This manual is related to the following products: