BCM567X BOARDCOM [Broadcom Corporation.], BCM567X Datasheet - Page 18
![no-image](/images/no-image-200.jpg)
BCM567X
Manufacturer Part Number
BCM567X
Description
A Scalable Approach to Gigabit Ethernet Switch Design
Manufacturer
BOARDCOM [Broadcom Corporation.]
Datasheet
1.BCM567X.pdf
(28 pages)
BCM567x/BCM569x
White Paper
06/27/02
O
S
D
D
THER
YSTEM
ESIGN
ECISIONS
In addition to deciding among the various types of component interconnects, other decisions must be made for each type of
switch to be developed. The primary ones are outlined below.
F
S
IXED
YSTEMS
The designer needs to make design choices if the system is to have more than 12 ports. If there are 12 or fewer ports in the
system, the designer can simply use a BCM5691 as a standalone switch.
To achieve larger port densities, the BCM5690 can be combined with additional BCM5690s or BCM5670 switch fabrics. At
this point, the designer must decide if the switch should be optimized for cost or for performance.
If the designer decides that the system cost is the most important factor, a cascade interconnect for linking the chip
componentry should be chosen, and the BCM5690s will be linked without the benefit of the BCM5670 switch fabric in a
shared-bandwidth configuration.
If the designer decides that system performance is the most important factor, a matrix interconnect for linking the chip
componentry should be chosen. In addition, a BCM5670 (or BCM5671) switch fabric will likely be included in the
configuration so that packets are transmitted over a switched, rather than a shared, backplane.
S
S
TACKABLE
YSTEMS
Here, the designer must decide on a per-switch port count, then choose among the optional modules for the features and
capabilities to be activated.
For the highest-redundancy levels in the stack, designers are likely to choose a matrix interconnect; for highest performance,
a daisy-chain interconnect is recommended.
Which cabling medium to use is another decision to be made when building stackable switches. System vendors can choose
fiber cabling or various categories of copper cabling, and gigabit interface converters (GBICs) are available for converting
electrical copper signals to fiber-optics. Designers also have the option of using a backplane trace to connect the switches
to one another directly.
M
C
ODULAR
HASSIS
When designing these backbone devices, a switch vendor must decide how to balance the redundancy of the backplane for
resiliency and uptime with the performance of each slot in the system. For redundancy, designers can choose to connect
the two HiGig connections in each blade to two separate switch fabrics in a master/slave configuration. Alternatively, the
designer can connect each slot in a full mesh configuration, rather than using a switching fabric. This is a highly redundant
setup; however, it does cause some performance degradation on each slot. From a performance perspective, if the system
vendor elects to configure two switch fabrics in a master/slave setup, Broadcom recommends running both switch fabrics
concurrently, rather than leaving one to sit idle in standby mode for failover.
From a management perspective, in the chassis system, the switch vendor must decide whether to run a centralized
management blade or to distribute management capabilities across individual blades. The latter choice is often more
scalable, as switch makers can add more complex management features incrementally to the system as required.
Bro adco m C orp or atio n
Page
12 Other System Design Decisions
Document
567x_569x-WP100-R