Gigabit Routing

canoeornithologistΔίκτυα και Επικοινωνίες

26 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

115 εμφανίσεις

Gigabit Ethernet
기술








whlee@etri.re.kr


한국전자통신연구원

Gigabit Ethernet


Ethernet


Ancestor: ALOHA, Xerox


IEEE 802.3
표준
(

85):
Shared media access(CSMA/CD), 10
Mbps


Fast Ethernet


IEEE 802.3u
표준
(

95)


Ethernet


동일한

프레임

포멧
,
엑세스방식
,
관리대상


차이점


속도
(100
Mbps),
동축

미디어

지원

불가
, 1
또는

2
개의

리피터




가능
(
max cable
길이
: 100
m),

PHY signaling (100Base
-
T4 for
Cat
-
3/4, 100Base
-
TX for Cat
-
5, 100Base
-
FX for fiber),
FDX/Flow control operation, Autonegotiation


Gigabit Ethernet


IEEE 802.3z(

98.6)


Ethernet


동일한

프레임

포멧
,
엑세스방식
,
관리대상


차이점


속도
(1000
Mbps),
1
또는

2
개의

리피터

사용

가능
(
max cable
길이
:
100
m), Carrier
확장
(
Slot time: 512 octs), Packet bursting,

PHY
signaling (1000Base
-
X for fiber/STP, 1000Base
-
T for Cat
-
5),
FDX/Flow control operation, Autonegotiation


802.3
z(MAC)


802.3
z MAC

기능


Carrier Extension


Frame Bursting


Half Duplex
동작시만

적용


802.3
z extents full duplex (802.3x) operation to
1000Mbps


Gigabit Ethernet
망은

주로

Full duplex


사용될

전망


Carrier Extension


Interoperable with Ethernet and Fast Ethernet


min. frame size: 64 octs


cabling distance: 100 m


Frame format with carrier extension

Preamble SFD DA SA Type/ Data FCS Extension


Length

64
octs min

512
octs min

Collision window


Packet Bursting


bandwidth waste: (512
-
64) octs carrier extension


Carrier extension plus a burst of packets

(
Extension bits)

(
IPG)

(
IPG)

CSMA/CD Parameter
비교

Simulation Results


Environments


1000Mbps Network


15 Stations (100% Offered Load)


512 Bytes Slot times


200m Network Diameter

Packet Size

References

Packet Packing

Packet Bursting

64
Bytes

1500
Bytes

Workgroup
분포

90.68
Mbps

339.84
Mbps

286.75
Mbps

853.20
Mbps

853.20
Mbps

853.20
Mbps

562.19
Mbps

740.80
Mbps

700.32
Mbps

802.3
z(PHY)


Functional Elements



8
B/10B

Encoding/Decoding

1000
BASE
-
CX

1300
nm SMF

1000BASE
-
LX

UTP

1000BASE
-
T

Media Access Control (MAC)

Full Duplex and/or Half Duplex

Gigabit Media Independent Interface GMII

1000
BASE
-
T

Encoder/Decoder

850
nm FC

1000BASE
-
SX

Twinax or Quad

25m+

SMF

3km+

50
um or 62.5um

MMF

(200m
-

550m+)

UTP

100m

802.3
z PHY

802.3
ab PHY

1000
BASE
-
T


802.3
ab project

spun
-
off


from 802.3z in 97/3


Takes advantage of high density, high speed silicon


Transmits bi
-
directionally on 4 wire pairs to achieve
full duplex communication (250 Mbps per pair)


Uses 5 level PAM signaling


Employs Digital Signal Processing for :


Echo cancellation


NEXT cancellation


Digital adaptive equalization

Gigabit Ethernet Product


Switches


pure Gigabit Ethernet switch


Gigabit/Fast Ethernet switch


Uplink/downlink modules


NICs


Gigabit Ethernet router interfaces


Buffered distributor


Gigabit Ethernet Migration


Switch to Switch


100/1000
스위치

간의

1000
Mbps
연결


Switch to Server


어플리케이션과

File
서버의

고속

Access


Switched Fast Ethernet Backbone


Gigabit Ethernet
스위치를

이용한

Fast Ethernet
집선


Shared FDDI Backbone


고성능

데스크탑

연결

Switched Fast Ethernet Backbone

Buffered Distributor


Full
-
duplex, Multiport, Hub
-
like Device


1Gbps
급의

2


이상의

포트

수용


모든

입력

트래픽을

자신을

제외한

모든

연결된

링크로

송신

(
No Address Filtering)


IEEE802.3x (

97.3.) Flow Control
사용

-

To prevent
internal buffer congestion


Shared Bandwidth Domain
제공



CSMA/CD in a Box



송신



1


이상의

프레임

버퍼링

장애요소


변형된

MAC


Slot Time, Burst Length Timer



Cabling
-

Fiber, UTP Cat.5


서버와

NIC


성능

한계

(
일반적으로

100
-
200
M)


Multimedia Application


필요한

QOS
부재


대역폭

활용의

비효율성


모든

Packet


64
Byte
경우



120
Mbps


평균

200
-
500
Byte


300
-

400 Mbps


Packet Bursting


이용에

문제

(
어플리케이션
)

Gigabit Ethernet


Multimedia


High capacity and Low latency


IEEE802.1Q


VLAN Packet Header


Priority


IEEE802.1p


Multicast Group
지정

:
Packet Flooding
감소


Priority Queueing


RSVP, RTP, RTCP


Gigabit Ethernet Switch



Gigabit Multilayer
스위칭


기술

채택


Gigabit Link
+
Multilayer Switching + Gigabit
Performance(line rate, wirespeed)


Gigabit Link Technology


Fiber Channel


HIPPI, serial
-
HIPPI, superHIPPI(Gigabyte System Network)


Gigabit Ethernet


Gigabit Link Aggregation


Multilayer Switching


Layer 2 only: Ethernet
스위치
(
Multiport
브리징
)


Layer 3: Routing


route once, switch many technique: MPLS, MPOA, IP
Switching(Ipsilon), Netflow Switching(Cisco), Fast IP(3Com)


wirespeed routing technique: Hardware
-
based traditional router


eliminate the performance bottlenecks of the software
-
based
routers


improve routing performace by implementing specialized routing
ASICs and switching fabric


Layer 4: TCP/UDP port number(uniquely indentify
application protocols(e.q., HTTP, SMTP, FTP, etc))



Gigabit Performance


Wirespeed performance for Gigabit Ethernet link


switching and routing at a max. of 1.488 Mpps per port
(assuming all packets with 64 bytes in length)

Packet per sec

K M

86 97 2000


-

wirespeed routing
technique(Gigabit
routing switch)

Supply

Demand


Integration of Layer 2 and Layer 3 in Today

s LAN

Bridges

Routers

Ethernet

Switch

More bandwidth

10/100
M

Ethernet

Gigabit

Ethernet

More bandwidth

Layer 3

Switching

Gigabit

Routing

Switch

통합

More bandwidth

and

performance

at lower cost

More functionality

Ethernet Switching Technology


L2 switch


Forward frames based on the Layer 2 Ethernet, token
-
ring,
or FDDI MAC addresses


Increasing bandwidth without adding unnecessary complexity
to the network


desktop/workgroup complexity connectivity and network
segmentation


Two types


Workgroup switch: effectively replacing shared
-
media wiring
hubs


Segment switch: bridging traffic between LAN segments; each
port must support large numbers of MAC address


Switching as a Routing Alternatives


In the past, the role of router was interconnectivity between
LANs within the campus/building network


poor performance


congestion


Layer 2 switches are superior to the past routers


Higher throughput and lower latency


Lower cost per port: 3 to 10 times the price per port


Easier administration: plug and play



Today

s network: widespread adoption of LAN switching



Problems in Today

s

switch
-
centric


network


Poor scalablity


Broadcast storms


Spanning tree loops


Inefficient addressing limitation


Broadcast impact on workstaions CPU utilization



VLAN or Routers can solve the above problems

Î
Shortcomings: Higher cost, worse performance, and greater
administration



Why routing?


Traffic filtering for WAN links


Broadcast storm avoidance


Multiple Administrative domains


Security


Routing

s traditional shorcomings


Poor routing performance: <
수백

Kpps


High latency



Routing performance is the major bottleneck in today

s
network



Switch when you can, route when you must
´

slogan

Typical Switched Network


One
-
armed Router


Router

Layer 2

Switch

Layer 2

Switch

Layer 2

Switch

WAN

Bottleneck

Pumping

Gigabit Traffic

20%
Intra
-
LAN

80%
Inter
-
LAN

New Generation of Routers


Needs


Routing performance is required more than ever according
to changing 80/20 rule


Support many time
-
sensitive multimedia applications


Support Gigabit speed topology and low latency


Require the benefits of using Layer 2 switching(e.q., high
throughput, low latency, lower cost per port)



A solution to this needs: Layer 3 switching


Two alternatives: The one is

Router Once, Switch Many
´

The other is

³
Gigabit Routing
´

Review of Routing Function


Route calculation


Identify the next hop for the traffic to traverse through the
network


Exchange the overall network topology information using
the routing protocol(RIP, OSPF, BGP,

)


Traffic filtering


Frame forwarding


Forward the input packet to the destination link at Layer 3



Layer 3 switching decople the forwarding from the routing
function to combine Layer 2 switching and the frame
forwarding


Mind
-
boggling Terminology


Cut
-
through routing


Switched routing, Routing switch, Switching Router


Layer 3 switching


Multilayer switching


Tag switching


IP switching


Multiprotocol over ATM(MPOA)


Multiprotocol label swapping(MPLS)



describe various techniques for optimizing network switching
performance


Route Once, Switch Many



Flow
-
based routing


Combine Layer 2 switching and Layer 3 packet
forwarding


Improving routing performance


Addressing WAN scalability


Many vendor architecture


Ipsilon IP Switching, Toshiba Cell Switch Router: IP/ATM
solution


Multiprotocol over ATM(MPOA): IP/ATM solution


MPLS


3Com FastIP, Cisco Netflow Switching


Gigabit Routing



Standard packet
-
by
-
packet routing


Improving routing performance by hardware
-
based
packet forwarding engine and switching fabric


장점


Enhanced performance: packet routing at switching speed


Simplified management:
기존망과의

완벽한

호환성
,
관리의

간편성


Lower cost:
포트당

라우터의

1/3 ~ 1/10
수준의

가격


Comparison of Layer 3 Switching Implementations

Gigabit Routing

Route Once, Switch Many

기존망과의

호환성


벤더별

독자적인

프로토콜
/
방식

패킷별

라우팅에의한

지연


연결

설정

지연

분산

모델


중앙

집중

모델


ATM
망에서

유리


Legacy

LAN
에서

유리


간단
,
신뢰성


복잡
,
서버의

병목현상
,

Single point of failure

Design Issues in Gigabit Routing


High
-
speed route lookup


Architecure of a router


Output queueing


Input queueing

High
-
speed Route Lookup


Address lookup


Layer 2 address(48
-
bit MAC, 28
-
bit VPI/VC): Flat structure,
limited scalability, simple direct lookup(exact matching);
lookup at gigabit wirespeed simple


Layer 3 address(IP, E.164): Hierarchical structure, complex
lookup(longest prefix matching); lookup at gigabit
wirespeed difficult

Forwarding table

<network addr/mask, port>

<128.32.1.5/16, 1>

<128.32.225.0/18, 3>

<128.0.0.0/8, 5>

Destination

Address

128.32.195.1

match

Multiple matched

entry

<129.32.0.0/8, 2>

Longest
matching


Cost of

Layer 3 Route Lookup


Number of entry:
백본의

경우

수천개

이상


Multiple matching problem


Various destination:
백본의

경우

60


동안

64,000
개의

상이


목적지를

갖는

패킷


Frequency of small packet size: 70%
정도가

1000
bit
미만


Performance Improvement


Hardware oriented techniques: CAM, special ASIC


Table compaction techniques: compact table structure


Hashing techniques: scalable hash
-
based algorithm


Architecure of A Router


Generic structure

Routing


Processor

Line

Card

Route calculation/

frame forwarding
software

Line

Card



Switching

Fabric



Small

forwarding
cash

Line

Card

Line

Card

Input/

output

port

Input/

output

port


Switching fabric techniques


Shared bus: bus arbitration overhead


crossbar: scheduling speed limit


shared memory: memory access time limit


Switching scheme


cell
-
based switching: front
-
end segmentation and
reassembly module around ATM
-
cell switch core


variable
-
size packet switching: arbitration and schedule


Queueing scheme


Output queueing: fast switching, buffer access time


Input queueing: HOL blocking, switch fabric/output port
scheduling



Classification


Single processor
-
Shared Bus


Multiple procesrors
-
Shared Bus


Multiple processors
-

Space switching


Shared parallel processors
-
Space switching



fixed
-
size switching


variable
-
size switching



Output
-
queued swithing


Input
-
queued switching


High
-
speed router(1)

Routing


Processor

Line

Card

Route calculation

software

Line

Card



Switching

Fabric



Small

forwarding
cash

Forwarding

Engine

Forwarding

Engine

Input/

output

port

Large

forwarding
cash


High
-
speed router(2)

Routing


Processor

Route calculation

software

Line Card/

Forwarding
Engine



Switching

Fabric



Large

forwarding
cash

Input/

output

port

Line Card/

Forwarding
Engine

Line Card/

Forwarding
Engine

Input/

output

port

Line Card/

Forwarding
Engine

Output Queueing


Speed
-
up techniques


Wide
-
width memory and parallel access


Specific memory read/write control logic


Queue scheduling


First
-
come First
-
served(FCFS)


Fair queueing


Weighted fair queueing

Input Queueing


HOL blocking problem


Arbitrating access to the switch fabric at high speed


Packet scheduling algorithm according to the
different speed of output links

Layer 4 Switching


제품차별화를

위하여

Gigabit router
벤더들에

의하여


들어진

용어


Definition


The ability to make forwarding decisions based not just on
the MAC address(Layer 2 bridging) or IP addresses(Layer 3
routing), but on the TCP/UDP(Layer 4) application port
numbers


Implementations


트래픽의

스위칭



포워딩시에

응용

서비스에

따른

우선

순위
처리를

가능케





outbound queuing, traffic prioritization, bandwidth
reservation mechanism


Example products


Quality of Service: provide different service quality levels
based on the type of application


L4

Routing

: change the network path taken by packets
based on TCP or UDP port number


Server Load Balacing: balance and direct the prioritized
traffic to the appropriate server


Problem


Large table capacity for wirespeed forwarding of traffic at
Layer 4


결론


Will switching kill routing?


수요전망