Documente Academic
Documente Profesional
Documente Cultură
Technical Overview.......................................................................1
High Reliability.......................................................................2
High Performance..................................................................4
Expansibility ..........................................................................4
IRF Distributed Device Management .............................................4
High-Efficiency Configuration Management............................5
Software Version Upgrade .....................................................5
Hot Swap of Devices .............................................................5
Distributed Layer 2 Protocol ..........................................................6
Distributed Forwarding ..................................................................7
Layer 2 Forwarding................................................................7
Layer 3 Forwarding................................................................8
Multicast Forwarding..............................................................9
Conclusion..................................................................................11
IRF
TECHNOLOGY
IRF Technical Overview White Paper
Technical Overview
IRF, Intelligent Resilient Framework launched by Huawei-3com is an innovational technology for
building core networks. It would help users to design and implement gigabit Ethernet core and
convergence backbone with high availability as well as high scalability.
IRF technology allows devise gigabit layer 3 switches to interconnect, forming a distributed
switching architecture, and run as one logic switching entity. For management and configuration, a
distributed switching schema acts like one switching device; for performance, every switch in it can
make local forwarding decision for second or third layer stream communications services of their
ports. Thus IRF provides users with a new stacking technology.
The devices that support IRF interconnect to form” a united equipment” called Fabric, of which each
device is called a Unit. Devise Units composing a Fabric becomes an integer whether in managing
or using. That is to say, users can manage and use the devices as single equipment. In this way, not
only can port numbers and switching capability get expanded by adding devices, but also reliability
of the whole Fabric can be improved through devise units backing up for one another.
Host Server
UNIT 1 UNIT 2
Host
Core Switch
UNIT 3 UNIT 4
IRF FABRIC
Host
Link Aggregation
Common Link
Stack Link
Compared with traditional stacking technology, IRF is a more enhanced stacking technology, which
gets innovation or enhancement in many aspects: Besides the capability of expanding ports and
unified management, IRF makes great improvement in aspects of high reliability as well as
redundancy backup. IRF technology allows for cross-device link aggregation in a global range,
providing an all-around link level protection. Meanwhile, IRF technology provides cross-device layer
3 routing redundancy, which supports diverse Unicast Routing Protocols, Multicast Routing
Protocols and distributed process, and really achieves multiple routing protocols hot standby
technology. For a long time, hot standby technology for Unicast Routing Protocol and Multicast
Routing Protocol has been a specific technological barrier in the field. IRF technology, however,
appears and puts forward a completely new solution to high reliability. Furthermore, IRF technology
realizes distributed operation of layer 2 protocols in a Fabric, improving unit utilization and reliability,
reducing protocol dependency among devices.
To be specific, IRF mainly consists of three technical aspects: DDM, DRR and DLA.
n DDM (distributed device management): seen form outside, the whole Fabric is single
equipment. It allows users to manage the Fabric through diverse modes: CONSOLE,
SNMP, TELNET, WEB, etc.
n DRR (distributed resilient routing): For users, various devices of a fabric seem to be a
single layer 3 switch. The whole Fabric will, as single equipment, perform routing
function as and forwarding function between layer 2 and layer 3. As Unicast Routing
Protocol and Multicast Routing Protocol run in distributed mode and entirely support
hot backup, when there occurs fault in a certain device, routing protocols and data
forwarding can continue.
n DLA (distributed link aggregation): it supports cross-device link aggregation, capable
of link load balancing and mutual backup among devices.
High Reliability
I. Redundancy Backup
Compared to traditional device networking, IRF provides real Unicast Routing Protocol and Multicast
Routing Protocol hot standby. And users needn’t spend half of their investment on backup, for all
devices in IRF actually take part in service running. IRF performs backup simultaneously as it
provides services.
DRR realizes routing protocol hot standby technology, achieving strict synchronization of routing
information for every unit in Fabric, besides, of which when one or more units fail, other units run as
usual and take over function of the failed unit immediately when IP route in field won’t be accordingly
interrupted, nor would forwarding stream as well as services of layer 2 and layer 3, thereby really
realizing failure protection and device switch functions without halting routing protocols and
services.
Distributed aggregation technology continually eliminates the problem of aggregation devices single
point failure and improves usability for aggregation link. As aggregation members are allowed to lie
in different devices in fabric, even if the whole device that some members lie in crashes, aggregation
link won’t be entirely invalidated. Other normal units would continue to manage and maintain the rest
aggregation ports. That has great significance for core switch system and network condition that
requires high quality services.
Datagram
IRF FABRIC
Datagram
Link
IRF FABRIC
Failure
DLA technology allows other devices out of IRF network core, like switches, to get multi-homed
access to IRF network core, which greatly improves the usability of the whole network. Stream into
IRF network core is equally distributed in aggregation links. When an aggregation link crashes, DLA
is capable of automatically redistributing stream to rest aggregation links to realize resilient backup
for links and enhance network reliability.
High Performance
Because IRF equipments are stacked by many stand-alone devices that support IRF features,
switching capacity and port numbers of IRF equipment are the sums of those of stand-alone devices
in IRF. Hence, IRF technology has the ability to easily expand user port density and device core
switching capacity for several times, thereby greatly improves performance of the equipment.
Expansibility
IRF technology supports stack of eight devices at most. Users could purchase IRF switches
according their needs. If users have to expand network capacity owing to services request, what
they need to do is only to stack more IRF switches. IRF technology greatly protects users’ previous
investment and thus enhances users’ investment efficiency.
IRF allows as many as eight devices to form a Fabric, and no matter from management feature or
from forwarding feature, to users, the Fabric runs as one device. Every device of a Fabric shares the
same bridge MAC address, single manage IP address and layer 3 forwarding address. No matter by
what kind of modes (CONSOLE port, TELNET, or SNMP) a user access any of the devices within a
Fabric, he needs to configure the Fabric only once, and accordingly all the devices are sure to be
configured.
For previous stack technology of simple stack and multiple devices, software upgrade action has to
be implemented repeatedly, while for IRF switches, only once by Web network management,
software of all units in the Fabric will be automatically upgraded.
Devices in a Fabric can be inserted in and draw out randomly, without any effect on current normal
running of the Fabric.
When a device is connected to a running Fabric through the stacking wire, the original Fabric will
detect and authenticate the new device whether it can be added. If it can, it will receive original
global configuration and forwarding information, meanwhile spread its unique configuration as well
as forwarding information into the Fabric. After transitory information transaction, the new device will
be capable of participating in forwarding. Moreover, the procedure makes no, even a little, effect on
forwarding process of original devices. Two existing Fabric also can be merged, and respective
forwarding won’t be affected.
When one or more units are disconnected from a Fabric, related configuration and forwarding will be
deleted out of the rest devices, thereby making no forwarding failure.
At the same time, the breakaway units are inclined to form a new Fabric. If initial configuration has
layer 3 interface, IP address of the new Fabric will collide with that of original Fabric, or different
devices possess the same IP address in a network. To avoid the case, IRF adopts Resilient Arp
technology to check IP address collision between original Fabric and breakaway units, if there is a
collision, one of them will be degraded to layer 2 device in order to avoid network route oscillation
caused by collision.
Host2
Unit4 Unit3
NORMAL STACK
Host3
Host4
Each unit of IGMP-snooping that runs in IRF switches just maintains its own router port and host port,
neglecting port numbers of other units. If the device receives IGMP host join packet, which indicating
it needs to forward multicast data, the unit will send a piece of information to other units in the fabric,
informing them of forwarding data to it. In case of a second reception of host join packet from the
same multicast group, the device will not inform other units again. Forwarding IGMP packets as
needed decreases congest collision of IGMP protocol packets in the Fabric.
Host2
Unit4 Unit3
IRF FABRIC
Host3
Host4
Distributed Forwarding
The devices in the fabric are connected with each other through stack ports, and there are various
kinds of connection modes for stack, such as serial connection, ring connection and star connection.
If ring connection is adopted and packets need to be forwarded from other units in the fabric, then
unit that receives packets will have two routes to forward the packets to unit in which egress port
resides. When IRF performs forwarding, it will choose a routing, which has the closest distance off
the above unit. Compared with unidirectional forwarding, this forwarding mode is effective and has
function of load balancing.
IRF technology thoroughly realizes distributed forwarding of packets. And distributed forwarding can
be implemented both in switch of Layer 2 packets and in routing of Layer 3 packets. Distributed
Forwarding utilizes bandwidth of units in the fabric to the utmost.
Layer 2 Forwarding
In every unit of a Fabric there are adequate layer 2 forwarding entries, which guides packets to finish
switch locally, with no need to be transacted by a third party. When a unit in the Fabric receives
forwarding packets, through scanning its layer 2 forwarding form the unit can get the forwarding
egress port, which can be either local ports or ports in other units. If egress port resides in the unit,
packets will be switched out direct; if it resides in other units, packets will be forwarded to
corresponding units by stack ports and then switched out. Anyway, it needs to query layer 2
forwarding form only once for switch whether the port resides in local unit or other units.
Any unit in the Fabric, after receiving forwarding packets, will learn source MAC address as any
other switch does; and no matter how many devices in the Fabric there are, IRF switch must behave
as one switch so that layer 2 forwarding entries learned by a unit would be possessed by other units
in the Fabric, otherwise packets will be broadcasted in VLAN. To realize layer 2 distributed
forwarding in the Fabric and make switches in IRF behave as one device, when a unit has learned
new MAC entries and those users configured, it has to synchronize the entries to other devices in
the Fabric. When the Fabric changes (for example, adding a new unit), the new unit has to obtain
forwarding entries of original Fabric and meanwhile synchronizes its own ones to the Fabric. If a unit
is disconnected, entries related to the unit would be deleted from the Fabric.
Layer 3 Forwarding
Similar to layer 2 forwarding, IRF switches realize distributed layer 3 forwarding, that is to say, any
unit in a Fabric has the function of complete layer 3 forwarding: when it receives layer 3 packets to
forward, by scanning its layer 3 forwarding form the unit can get the forwarding egress port as well as
next hop, and then sends the packets through the right egress port, which can be either local ports or
ports in other units. Since the Fabric works as one device throughout the procedure, the port of any
unit is the port of the Fabric, so that sending packets from one unit to another becomes pure internal
implementation, which is shielded from external. It means that, for layer 3 packets, the hop number
increases merely one no matter how many units the packets go through in the Fabric, or they seems
through only one network device, and internal transfer of packets won’t change their layer 3
characters. Therefore, for external devices, the Fabric is one device all the time.
Router1
IP Packet Data Flow
Unit1 Unit2
Router2
Unit4 Unit3
IRF FABRIC
Router3
Router4
Whether the forwarding egress port resides in the local or other devices in the Fabric, packet layer 3
forwarding can be realized through scanning local routing forwarding form only once. Compared with
centralized forwarding, It reduces the dependence on one device and decreases forwarding load for
devices.
Multicast Forwarding
Multicast data forwarding is the key cause of affecting network bandwidth, and how to reasonably
and efficiently deal with multicast data forwarding becomes a troubling issue in the field.
Compared with traditional stack technology and equipment interconnected by multiple stand-alones,
devices stacked through IRF technology are capable of forwarding multicast data packets as needed,
thereby decreasing multicast data stream between devices.
IRF switches, which organically combine layer 3 multicast forwarding form with IGMP-snooping,
produces layer 2 and layer 3 combining multicast data forwarding form which guides multicast data
forwarding in switching. The multicast forwarding form only maintains the user egress port in local
unit and other units’ egress ports in the Fabric. This highly efficient forwarding form radically solves
the problem of multicast data occupying bandwidth. Thus it assures there is only one copy of
packets transferring between devices.
Source Host 1
Host2 Host 2
Host 3
Host 4
Host 5
Host3
Host4
Source Host 5
Host2 Unit 1
Unit 4
Forwarding Table
on Unit2
Unit4 Unit3
IRF FABRIC
Host3
Host4
Conclusion
IRF technology is a distributed network technology. It enables network designers to build the
network core with high reliability as well as flexibility. Its performance, configuration capacity and
flexibility have the ability to increase with network synchronously. Thus it avoids the problems as
one-off larger investment and physical limitations that centralized network core devices are
confronted with.