Brief
SANDBURST® SWITCH FABRIC
®
HIGH-END PACKET SWITCH FABRIC FOR METRO AND
ENTERPRISE CORE NETWORKING SYSTEMS
Sandburst Fabric Block Diagram
•Glob al Arbitrat ion
•Patented global arbiter provides the best scheduling
performance; the scheduler has global state information for
every queu e and port in the syst em.
•Virtual Output Queuing (VOQ)
•Supports up to 16K Virtual Output Queues for efficient
scheduling.
•Single-Stage Buffering
•Single-stage buffering on ingress only.
•No packet reassembly or reordering required.
•Up to 16 Classes of Service (COS)
•Support for up to 16 AF classes of service.
•Three levels of drop precedence per COS.
•Flow-Based Queuing on the Egress
•Flow-based queuing on the egress with support of up to 16K
virtua l po r ts .
•Up to eight level s of hiera rc hy.
•Scalable Architecture
•Fabric scales from 20 Gbps to 1.28 Tbps.
•Non-Bloc ki ng Pe rf or ma nc e
•Line-rate operation under all packet sizes and traffic patterns
for unicast and multicast traffic .
•No head of line blocking.
•Advanced Quality of Service (QoS)
•Supports Service Level Agreements (SLAs) and bandwidth
guarantees.
•Maintains priority and classes of service through the fabric.
•Egress queuing per virtual port.
•Reliability
•Fault tolerance and redundan cy features.
•Error detection and correction mechanisms.
•ECC memory protection in all devices in silicon, and CRC
protection at the system level.
•Metro-Ethernet Core and Edge Switch-Routers
•Enterprise Core and Wiring Closet Switch-Routers
•IP Routers
•Multi-Service Provisioning Platforms
KEY ARCHITECTURAL ADVANTAGES SUMMARY OF BENEFITS
APPLICATIONS
XXXXX
Packet Processor
Media
Interface
SPI-4.2
TME-2000
Queuing Engine
SPI-4.2
QE-2000
Queuing Engine
Packet
Buffer
Switch Fabric
Interface
Switch Control
Interface
Switch Fabric
SE-4000
BME-3200
Bandwidth
Management
Engine
Fabric Card
OVERVIEW
®
Phone: 949-926-5000
Fax: 949-926-7761
E-mail: info@broadcom.com
Web: www.broadcom.com
BROADCOM CORPORATION
5300 California Avenue
Irvine, California 92617
© 2008 by BROADCOM CORPORATION. All rights reserved.
Sandburst_Switch_Fabric-PB01-R 02/20/08
Broadcom®, the pulse logo, Connecting everything®, Sandburst®, and the Connecting everything logo
are among the trademarks of Broadcom Corporation and/or its affiliates in the United States, certain
other countries and/or the EU. Any other trademarks or trade names mentioned are the property of their
respective owners.
Ultra-Scalability
The San dburst® switch fabric is a highly scalable, non-blocking packet
fabric that s cal es from 20 Gbps to 1.28 Tbps wi th no performance
degra dation, regard less of QoS p rofiles or tra ffic m ix. D esigne d to m eet
the rigorous requirements of both the Enterprise Core and
Metr o-Ethern et markets , t he switch fabric de livers ba ndwidth
guarantees, advanced traffi c engineering , an d per-flow queui ng across
its extremely low-latency and highly-resilient fabric.
Optimal Traffic Management
The switch fabric delivers these service s using a patented Virtu al Output
Queued (VoQ) architecture. The architecture supports up to 16K queues
to enable over 16 classes of service per physical port or physical queuing
per customer for serv ice provider applicat i ons. The switch fabric
optimally sched ules traffic acro ss its fabric using a global view of
bandwidth requi rements. Its patented global sche duling algorithms
enable support of highly-efficient and granular bandwidth guarantees,
and accurat e prioritization of fl ows through the fabri c.
Failover Schemes
The Sandburst switch fabric offers a rich selection of hardware failover
schemes for its system OEM cu stomers. The fabric supports 1+1, and
N:N graceful redundancy schemes with sub-50 ms failover for control
and data traffic types.
Advanc ed QoS
All of the switch fabric devices (see column to the right) provide
sophisticated mechanisms for QoS management through the fabric and
on the egress, a highly desired feature in the Metro-Ethernet Networking
market. The devices fully support the IETF DiffServ Per Hop Behaviour
(PHB) traffic management scheme supporting all the IETF-defined
PHBs; Assured Forwarding (RFC 2597), Expedited Forwarding (RFC
2598), and Best Effort. In add ition, packets can be queue d per customer
on the egress and hierarchicall y managed. The QE-2000 supports up to
eight levels of hierarchy.
Additional Egress Queuing Functionality
Systems designers have the flexibility of using the egress queuing
functiona lity in the QE-2000 a long with Virtual Output Queu ing. This is
achieved by using the QE-2000 in mixed mo de and dyna mically
partitioning memory buffers into VoQs and egress queues. In egress
mode, the entire 16K queues can be used for hierarchical egress queuing.
Silicon Devices
Que uing Engine (QE -2000)
•20 Gbps full-duplex traffic manager
•Manages up to 16K queues
•Supports up to 16 C lasses of Serv ice (CoS)
•Deep ingress buffers for Round T rip T ime (R T T) and QoS—using up
to 512 MB of ex ternal commodity DRAMs
•Strict Priority (SP) and Weighted Fair Queuing (WFQ) scheduling
•Fully suppor ts the IETF DiffServ model
•BE, EF, AF IETF PHB
•Per-flow egr ess queuing with up to eight leve ls of heirarchy
•2 x SPI-4.2 (OIF) 48/64 channels per SPI-4.2 interface
•PCI—32-bit, 66 MHz
Bandwidth Management Engine (BME-3200)
The BME-3200 is the second generati on global bandwid th
managerelement of the Scalable Ethernet Switching product line. The
BME-3200 connects to each of the QE-2 000s in a system, offering
highly reliable N+1 operation. The BME-3200 enables a highly scalable,
non-blocking packet fabric thaqt scales from 20 Gbps to 1.28 Tbps with
no performance degradation, regardless of QoS profiles or traffic mix.
The BME-32 00 is designed t o operate and me et the vigorous
requirements of both the Enterprise Core and Metro-Ethernet markets. In
tandem with the QE-2000 and SE-4000 it forms part of the switch fabric
which delivers bandwidth guarantees, advanced traffic engineering, and
per-flow queuing across its extremely low-latency and highly-resilient
fabric.
•Patented VOQ architecture
•Administers IP SLAs within the packet fabric
•Arbitra t es up to 32 QE- 2 00 0s us in g a de dicate d out-o f - band
Bandwid th Management Interface to communicate wi th each
QE-2000
•Dynamically allocates bandwidth to queues to provide bandwidth
guarantees
•Globally tracks VOQ leng ths across all QE-2000s
•Integrated WRED engine that calculates multiple-drop probabilities
for each CoS in the system
Switchin g E ngi ne (SE-4000 )
•40 x 40 cross-point switch
•Over 40 in. repeaterless oper ation of FR4
•Self-routing cross-connect
•Spatial multi casting
•Time tolerance architecture with Plesio-sy nchronous interfaces,
reducing clock complexity and cost
•Hardware-based link and system monitoring
