Managing 10GbE migration
http://www.cieonline.co.uk/cie2/articlen.asp?pid=1 [2008-6-4]
Tag: filtering fabric
Managing 10GbE migration Fabric demands and migration paths to 10 GbE and how to tackle themwith a modular hub.
While most applications servers such as the processing of sessionsor gateways are way below the need for 10GbE interfaces, processingof IP media may need 10GbE high-speed uplinks to interconnectsystems and to distribute traffic to the processing blades withinthe system. Another key requirement is the operation of multiplenetworks within the backplane. Future generations of processorblades may connect over 10GbE on the backplane. A modular design ofthe Kontron AT890x ATCA hub allows easy upgrades and growthaccording to demand. The paper shows a summary of fabric demands,migration paths to 10 GbE and how to tackle them with a modularhub.
Uplink and fabric capacity in ATCA
For telecommunication systems, there is huge diversity ofapplications. Concerning their systems architecture, communicationservers can be classified into the following major categories:
(1) Session Processors , such as Call Servers, Media Gateway Controllers, or IMS,
(2) Gateways , i.e. systems with TDM interfaces (Time Division Multiplex) toconnect circuit switched telephone networks or radio base stationsand convert signalling and traffic between TDM and IP,
(3) Media Processors to support media streaming and transcoding of media for contentadaptation, content filtering or IPTV,
(4) Packet Processors for deep packet inspection and filtering at packet level.
For the majority of applications, the total traffic per system willstay well below 1 GbE. This is the case for Session Processors,Gateways and Media Servers in mobile networks. Media Processor forIPTV (over DSL or Cable TV networks) and Packet Processors may needhigher uplink capacity of 10 GbE to interconnect with other systemsand to distribute traffic to the processor blades.
The system contains different kinds of processing elements: Sessionprocessors blades to handle media sessions, media processor bladesincluding DSPs for the conversion of media, and main controllers torun the system. The system configuration is redundant. There aretwo hubs to pick-up and distribute traffic. Each hub contains onemain controller, with one of them active and another one instandby. The Session and Media Processors operate in load sharingand are connected to both hubs over independent networks.
In the basic configuration, each hub may pick up 4x GbE of trafficfor the Base interface, and 4x GbE of traffic for the Fabricinterface via its front panel uplinks. Port aggregation may be usedto bind those links together. The Hub blade comprises separateswitches for Fabric and Base in order to keep the user plane andcontrol plane separated. Thus, the system contains 2 fullyredundant and completely separated networks in the backplane (whichrepresents the standard ATCA architecture). The AT8902 hub allowsrunning a 3rd network in the backplane over the Fabric interface.This network is also handled by the fabric switch. In order to keepthe networks separated, the fabric switch supports VLANs andprotected port groups. The same mechanisms may be used for furthersegmentation of the networks.
In total, each processor blade may connect over 6x GbE, with 3x GbEto each hub. With each hub able to pick up 8x GbE, the volume oftraffic is balanced for a 6-Slot system with 4 processor blades.For a larger system, e.g. with 14 Slots and 12 processor blades, ahigher uplink capacity may be desirable. Also, several systems maybe interconnected to each other and need extra capacity at theuplinks to handle interconnection traffic. So how to upgrade to10GbE?
How to handle growth?
The ability to upgrade 1GbE uplinks or 10GbE uplinks certainlyfacilitates handling growth. A first step is the provision of 2x10GbE Fabric uplinks per hub. Should there may be higher demands oninterconnection traffic, it should be possible to provide extra 2x10GbE uplinks per hub. A bigger step of migration is to move to aFabric of 10 GbE in the backplane. Such a step could happentogether with upgrade of 1GbE processor blades to 10GbE processorblades.
For a 14-Slot system with 12 processor blades and 4x GbEconnectivity to each slot and high traffic load for Media Server,an uplink capacity of 10 GbE per hub seems to be adequate. An extra10 GbE uplink may be used to interconnect systems. A desirableapproach to handle such upgrades is to add uplink capacity byadding modules. Should higher uplink capacities be needed, the hubsshould be capable to accommodate extra 2x 10GbE uplinks and thecorresponding switching capacity.
A more difficult step is the migration to 10GbE in the backplane.The difficult part is that most of todays processor blades operateat 1GbE and it is difficult to mix 10GbE blades and 1GbE bladeswithin the same backplane and over the same hub. Should individualblades, such as I/O processing blades or packet processing bladesneed to communicate with others over 10GbE, one possibleintermediate step is to use a meshed backplane and operate the10GbE in a point-to-point mode. Alternatively, the complete systemneeds to move to 10 GbE. While the backplane can operate with 10GbEover 4 Fabric ports (instead of individual Fabric ports for 1 GbE),the implications are expensive replacements for hubs and processingblades. While an upgrade of backplane capacity usually also impliesan upgrade of the total system capacity and processing power (i.e.new node boards), the replacement of hubs is surely wasted. Aconcept, which also allows upgrading the hubs to a 10GbE Fabric,thus represents a serious contribution to protect investments inapplications where 10GbE may be an option later on.
AT890x Modular Design protects investment
The AT890x hub series is based on a modular concept. In order toease the selection of the right configuration, it is available inthe following packages: the AT8901 for Base interface (GbE) only,the AT8902 with additional GbE Fabric interface, and the AT8904with a 10GbE Fabric interface. Each configuration supports 14 and16 slot shelves and each configuration contains 2 AMC slots forcustomisation. The AT8902 provides uplinks for the Base and Fabricinterfaces on the front panels with 4x GbE. All configurationscontain a Kontron IPMI controller (IPMC) implementation and supporthot-swap capabilities and full redundancy.
The Base interface in handled by a separate, non-blocking layer 2/3switch with 2x GbE to the backplane and 1x GbE connection to eachAMC slot. Among the routing protocols are OSPFv2, RIPv2, VRRP,DiffServ and ARP. The Fabric interface is handled by anothernon-blocking layer 2 switch with support of VLANs and protectedport groups. It provides 1x GbE to up to 16 slots, and extra 1x GbEconnectivity to the logical slots 2, 3, 4 and 5. It also provides2x 10GbE connectivity to each of the AMC slots.
The two AMC slot may be used for customisation, for example to host
" Processor AMCs as main controllers
" Storage AMCs as mass storage devices for Processor AMCs
" Dual 10GbE uplink AMCs to interlink the Fabric.
The AM4310 Interlink AMC provides 2x 10GbE uplinks on the frontpanel via optical XFP connectors and CX4 connectors, which itroutes via XAUI ports to the AMC connectors (using AMC ports 4-7and 8-11). The AT8902 hub picks up the 10 GbE ports and switchesthem to the Fabric interface. The AM4310 Interlink AMCs representone part of the modular design of the AT890x hub.
Fabric mezzanine
The second essential module is the Fabric mezzanine on the AT890x.The fabric mezzanine determines the type of AT890x package: a GbEFabric mezzanine generates the AT8902 for a PICMG 3.1 GbE Fabricinterface including customisation with extra 10 GbE uplinks withthe AM4310. Using the same platform, a different Fabric mezzaninecreates the AT8904 with a PICMG 3.1 10GbE Fabric interface, whilemaintaining the customisation with 2x 10GbE to both AMC slots for10GbE uplinks. The 10GbE Fabric with 10GbE connectivity to up to 16slots is also handled by a non-blocking layer 2 switch includingsupport of VLANs. The mezzanine concept also allows providing otherPICMG3.x extensions, such as Infiniband, RapidIO or ASI.
The AT890x provides managed switches with access over CLI, SNMP andTELNET. It supports IPMI version 1.5. Among the protocols supportedat Ethernet layer are link aggregation (IEEE 802.3ad), VLANs(802.1Q), Spanning Tree (802.1D), QoS (802.1p), flow control(802.3x), GVRP and GMRP. At IP layer, it supports OSPFv2, RIPv2,VRRP, IGMP snooping, IPv4 forwarding, DiffServ, ARP and ICMP.
The mainstream version of the AT890x provides a high degree offlexibility to cover different demands of functionality andcapacity. Its modular design enables existing installations to beupgraded rather than to enforce replacements of the whole hub. Itthus significantly contributes to protect investment. In order tosupport cost critical projects at high volume, the AT890x is alsoavailable in light versions with a reduced level of functions andcomponents.
In summary, Media Servers beyond mobile networks will need tohandle traffic demands beyond 1GbE. While link aggregation allowscovering increasing traffic demands to some extent, the next stepis to pick up traffic over 10 GbE uplinks, and interconnect withother systems over 10 GbE. In a multiprocessor system, the trafficcan be distributed to processing nodes over 1GbE connections.Another key requirement is the capability to run multiple separatednetworks within the same system in redundant configurations. TheAT8902 hub covers both: it supports easy upgrades with 10GBEuplinks via AMCs and can handle multiple networks within thesystem. The AT8902 also allows Fabric upgrades to 10GbE byexchanging its Fabric mezzanine. Thus, it represents the rightchoice to handle growth.
Kontron
www.kontron.com
Dr. Stephan Rupp is Senior Systems Architect and Wolfgang H
Managing 10GbE migration Fabric demands and migration paths to 10 GbE and how to tackle themwith a modular hub.
While most applications servers such as the processing of sessionsor gateways are way below the need for 10GbE interfaces, processingof IP media may need 10GbE high-speed uplinks to interconnectsystems and to distribute traffic to the processing blades withinthe system. Another key requirement is the operation of multiplenetworks within the backplane. Future generations of processorblades may connect over 10GbE on the backplane. A modular design ofthe Kontron AT890x ATCA hub allows easy upgrades and growthaccording to demand. The paper shows a summary of fabric demands,migration paths to 10 GbE and how to tackle them with a modularhub.
Uplink and fabric capacity in ATCA
For telecommunication systems, there is huge diversity ofapplications. Concerning their systems architecture, communicationservers can be classified into the following major categories:
(1) Session Processors , such as Call Servers, Media Gateway Controllers, or IMS,
(2) Gateways , i.e. systems with TDM interfaces (Time Division Multiplex) toconnect circuit switched telephone networks or radio base stationsand convert signalling and traffic between TDM and IP,
(3) Media Processors to support media streaming and transcoding of media for contentadaptation, content filtering or IPTV,
(4) Packet Processors for deep packet inspection and filtering at packet level.
For the majority of applications, the total traffic per system willstay well below 1 GbE. This is the case for Session Processors,Gateways and Media Servers in mobile networks. Media Processor forIPTV (over DSL or Cable TV networks) and Packet Processors may needhigher uplink capacity of 10 GbE to interconnect with other systemsand to distribute traffic to the processor blades.
The system contains different kinds of processing elements: Sessionprocessors blades to handle media sessions, media processor bladesincluding DSPs for the conversion of media, and main controllers torun the system. The system configuration is redundant. There aretwo hubs to pick-up and distribute traffic. Each hub contains onemain controller, with one of them active and another one instandby. The Session and Media Processors operate in load sharingand are connected to both hubs over independent networks.
In the basic configuration, each hub may pick up 4x GbE of trafficfor the Base interface, and 4x GbE of traffic for the Fabricinterface via its front panel uplinks. Port aggregation may be usedto bind those links together. The Hub blade comprises separateswitches for Fabric and Base in order to keep the user plane andcontrol plane separated. Thus, the system contains 2 fullyredundant and completely separated networks in the backplane (whichrepresents the standard ATCA architecture). The AT8902 hub allowsrunning a 3rd network in the backplane over the Fabric interface.This network is also handled by the fabric switch. In order to keepthe networks separated, the fabric switch supports VLANs andprotected port groups. The same mechanisms may be used for furthersegmentation of the networks.
In total, each processor blade may connect over 6x GbE, with 3x GbEto each hub. With each hub able to pick up 8x GbE, the volume oftraffic is balanced for a 6-Slot system with 4 processor blades.For a larger system, e.g. with 14 Slots and 12 processor blades, ahigher uplink capacity may be desirable. Also, several systems maybe interconnected to each other and need extra capacity at theuplinks to handle interconnection traffic. So how to upgrade to10GbE?
How to handle growth?
The ability to upgrade 1GbE uplinks or 10GbE uplinks certainlyfacilitates handling growth. A first step is the provision of 2x10GbE Fabric uplinks per hub. Should there may be higher demands oninterconnection traffic, it should be possible to provide extra 2x10GbE uplinks per hub. A bigger step of migration is to move to aFabric of 10 GbE in the backplane. Such a step could happentogether with upgrade of 1GbE processor blades to 10GbE processorblades.
For a 14-Slot system with 12 processor blades and 4x GbEconnectivity to each slot and high traffic load for Media Server,an uplink capacity of 10 GbE per hub seems to be adequate. An extra10 GbE uplink may be used to interconnect systems. A desirableapproach to handle such upgrades is to add uplink capacity byadding modules. Should higher uplink capacities be needed, the hubsshould be capable to accommodate extra 2x 10GbE uplinks and thecorresponding switching capacity.
A more difficult step is the migration to 10GbE in the backplane.The difficult part is that most of todays processor blades operateat 1GbE and it is difficult to mix 10GbE blades and 1GbE bladeswithin the same backplane and over the same hub. Should individualblades, such as I/O processing blades or packet processing bladesneed to communicate with others over 10GbE, one possibleintermediate step is to use a meshed backplane and operate the10GbE in a point-to-point mode. Alternatively, the complete systemneeds to move to 10 GbE. While the backplane can operate with 10GbEover 4 Fabric ports (instead of individual Fabric ports for 1 GbE),the implications are expensive replacements for hubs and processingblades. While an upgrade of backplane capacity usually also impliesan upgrade of the total system capacity and processing power (i.e.new node boards), the replacement of hubs is surely wasted. Aconcept, which also allows upgrading the hubs to a 10GbE Fabric,thus represents a serious contribution to protect investments inapplications where 10GbE may be an option later on.
AT890x Modular Design protects investment
The AT890x hub series is based on a modular concept. In order toease the selection of the right configuration, it is available inthe following packages: the AT8901 for Base interface (GbE) only,the AT8902 with additional GbE Fabric interface, and the AT8904with a 10GbE Fabric interface. Each configuration supports 14 and16 slot shelves and each configuration contains 2 AMC slots forcustomisation. The AT8902 provides uplinks for the Base and Fabricinterfaces on the front panels with 4x GbE. All configurationscontain a Kontron IPMI controller (IPMC) implementation and supporthot-swap capabilities and full redundancy.
The Base interface in handled by a separate, non-blocking layer 2/3switch with 2x GbE to the backplane and 1x GbE connection to eachAMC slot. Among the routing protocols are OSPFv2, RIPv2, VRRP,DiffServ and ARP. The Fabric interface is handled by anothernon-blocking layer 2 switch with support of VLANs and protectedport groups. It provides 1x GbE to up to 16 slots, and extra 1x GbEconnectivity to the logical slots 2, 3, 4 and 5. It also provides2x 10GbE connectivity to each of the AMC slots.
The two AMC slot may be used for customisation, for example to host
" Processor AMCs as main controllers
" Storage AMCs as mass storage devices for Processor AMCs
" Dual 10GbE uplink AMCs to interlink the Fabric.
The AM4310 Interlink AMC provides 2x 10GbE uplinks on the frontpanel via optical XFP connectors and CX4 connectors, which itroutes via XAUI ports to the AMC connectors (using AMC ports 4-7and 8-11). The AT8902 hub picks up the 10 GbE ports and switchesthem to the Fabric interface. The AM4310 Interlink AMCs representone part of the modular design of the AT890x hub.
Fabric mezzanine
The second essential module is the Fabric mezzanine on the AT890x.The fabric mezzanine determines the type of AT890x package: a GbEFabric mezzanine generates the AT8902 for a PICMG 3.1 GbE Fabricinterface including customisation with extra 10 GbE uplinks withthe AM4310. Using the same platform, a different Fabric mezzaninecreates the AT8904 with a PICMG 3.1 10GbE Fabric interface, whilemaintaining the customisation with 2x 10GbE to both AMC slots for10GbE uplinks. The 10GbE Fabric with 10GbE connectivity to up to 16slots is also handled by a non-blocking layer 2 switch includingsupport of VLANs. The mezzanine concept also allows providing otherPICMG3.x extensions, such as Infiniband, RapidIO or ASI.
The AT890x provides managed switches with access over CLI, SNMP andTELNET. It supports IPMI version 1.5. Among the protocols supportedat Ethernet layer are link aggregation (IEEE 802.3ad), VLANs(802.1Q), Spanning Tree (802.1D), QoS (802.1p), flow control(802.3x), GVRP and GMRP. At IP layer, it supports OSPFv2, RIPv2,VRRP, IGMP snooping, IPv4 forwarding, DiffServ, ARP and ICMP.
The mainstream version of the AT890x provides a high degree offlexibility to cover different demands of functionality andcapacity. Its modular design enables existing installations to beupgraded rather than to enforce replacements of the whole hub. Itthus significantly contributes to protect investment. In order tosupport cost critical projects at high volume, the AT890x is alsoavailable in light versions with a reduced level of functions andcomponents.
In summary, Media Servers beyond mobile networks will need tohandle traffic demands beyond 1GbE. While link aggregation allowscovering increasing traffic demands to some extent, the next stepis to pick up traffic over 10 GbE uplinks, and interconnect withother systems over 10 GbE. In a multiprocessor system, the trafficcan be distributed to processing nodes over 1GbE connections.Another key requirement is the capability to run multiple separatednetworks within the same system in redundant configurations. TheAT8902 hub covers both: it supports easy upgrades with 10GBEuplinks via AMCs and can handle multiple networks within thesystem. The AT8902 also allows Fabric upgrades to 10GbE byexchanging its Fabric mezzanine. Thus, it represents the rightchoice to handle growth.
Kontron
www.kontron.com
Dr. Stephan Rupp is Senior Systems Architect and Wolfgang H
Related News »
In Focus »
footwear exports
Last month, European footwear manufacturers proposed extending anti-dumping measures against ..
B2B Keywords:
International market Chinese Importer Wholesale trade Wholesale products World trade Wholesale distributors International trade Foreign trade Wholesale distributor Importers Import export business Sell online Help u sell Global trade How to market a product Online supplier Wholesale product
International market Chinese Importer Wholesale trade Wholesale products World trade Wholesale distributors International trade Foreign trade Wholesale distributor Importers Import export business Sell online Help u sell Global trade How to market a product Online supplier Wholesale product
Feature Products»
Hot Products: A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z | 0-9
-
-
About Us -
Links -
Privacy Policy -
Terms of Use -
Site Map -
Contact Us -
Call Center: 0086-95105561 © 2007 Tootoo.com -
