Transcript IP Switching and Gigabit Routers

IP Switching
and
Gigabit Routers
Shlomi Malki
Nachman Cohen
Jump to first page
Topics



Motivation.
Gigabit Routers.
IP Switching:
 flow
classification.
 implementation.
 GSMP/IFMP.

Conclusion.
2
Jump to first page
3
Jump to first page
4
Jump to first page
5
Jump to first page
Gigabit Routers





Multigigabit Routers.
IP/ATM.
Cell Switch Router (CSR).
IP Switching.
NetStart GigaRouter.
6
Jump to first page
Getting Up to gigabitRouter





Average packet size on internet is 2000
bits.
Therefore must forward 500 Kpps per
Gbps of traffic.
Replace shared bus by switch fabric.
Separate processor performs routing
function and scales with number of
peers. Not with bandwidth.
Increase forwarding performance with
multiple parallel forwarding engines.
7
Jump to first page
8
Jump to first page
Gigabit Router Components

Line Card
 Contains
the physical layer
components.

Switch Fabric
 Used
to interconnect the various
components of the gigabit router.
9
Jump to first page
Gigabit Router Components (1)

Forwarding Engine
 Inspects
packets headers.
 Determines to which outgoing
line card they should be sent.
 Rewrites the header.

Network Processor
 Runs
the routing protocols.
 Compute the routing table.
 Handle network management.
10
Jump to first page
ATM Overview






CELL Vs PACKET.
Segmentation & Reassemble.
Connection Oriented.
Virtual Channel, VCI.
Virtual Path, VPI.
Label Swapping.
11
Jump to first page
Switch Fabric


Offers much higher aggregate
capacity then the conventional
backplane bus.
Implementation:
 crossbar.
 ATM.
12
Jump to first page
Switch Fabric (1)

ATM advantage:
 Standard
H/W.
 QoS.
 Multicast.

ATM disadvantage:
 Cell
oriented.
 Connection oriented.
13
Jump to first page
Forwarding Engine

Location within the router:
 Physically
separate component.
 Integrated with either the line card
or the network processor.

At IP Switching most data need no
forwarding engine interference.
Whereas routers always requires
at least one forwarding engine.
14
Jump to first page
Design of the Forwarding Engine


As we already saw we must
forward 500 kpps per Gbps of
traffic.
Two approaches to achieve this
rate:
 the silicon forwarding engine.
 High speed general purpose
processor with destination
address caching.
15
Jump to first page
Silicon Design

Design


Silicon hardware.
Memory
16 byte for each IPv4 route table entry.
 250,000 routers.
 TOTAL: 4 Mbytes.


Forwarding capability
memory accesses per route = 1+logN.
 10 ns SRAM.
 200 ns for full lookup.
 TOTAL: 5 Mpps. ( enough for 10 Gbps )

16
Jump to first page
Processor with caching
Design

Design
A 415 MHz general purpose processor
with internal cache.
 Internal cache: least recently used of
9000 IPv4 destination address.


Memory


Additional external memory of 8 Mbytes
(holds the complete routing table).
Forwarding capability
11 Mpps - all requests are at cache.
 Multicast - handle by the full routing table.

17
Jump to first page
Design of the Forwarding Engine
Silicon Design
Processor with caching Design
A 415 MHz general purpose
processor with internal cache
Additional 8 MB (for a complete
Memory
4 MB
routing table of several hundred
routes)
Forwarding 5 Mpps on average 10 thousand
11 Mpps if
all the requested
Capability
Gbps of traffic
destinations in the cache
Maintains its
Maintains its full forwarding rate if
Advantage
maximum forwarding at least 60% chance the required
rate regardless past
stination
address has
in
Debate regarding
the been
use ofseen
caching
Disadvantage Fixed Solution
(locality).
Design
Silicon hardware
18
Jump to first page
Forwarding engine summarize



Sufficient to offer a simple, besteffort packets forwarding.
Additional functionality required of
the next generation of routers
(multicast, QoS differentiation,
firewall filtering, etc.)
Needs to base the routing decision
on more fields in the packets
header.
19
Jump to first page
IP Switching



Can used any higher level IP
functionality.
Uses the concept of a flow (a
sequence of packets that are
treated identically by possibly
complex routing function).
Uses an ATM switch as the switch
fabric.
20
Jump to first page
21
Jump to first page
ATM as Switch Fabric

The 3 approaches that uses ATM
as Switch Fabric are:
 IP/ATM.
 Cell
Switch Router (CSR).
 IP Switching.
22
Jump to first page
ATM as Switch Fabric

Incoming flows are mapped onto
ATM VC’s.
 The
IP Switch uses a protocol
IFMP (RFC1953) to propagate the
mapping between flow and VCI.
 IP/ATM uses a pool of preestablished PVC’s.
 CSR uses RSVP protocol
(RFC1577)to propagate the
mapping between flows and VCI’s.
23
Jump to first page
Flow Classification




Flow classification operation is to
select those flows that are to be
switched in the ATM switch and
those that should be forwarded in
the forwarding engine.
Long duration flows - ATM switch.
Multicast - ATM switch.
Short duration flows - Forwarding
engine.
24
Jump to first page
Flow Classification (1)
IP Switch Controller
IP Switch Controller
Po
rt
C
Port 1
Po
rt
C
Port 0
Port 1
Port 0
25
Jump to first page
Flow Classification (2)



For the flows selected for
switching, a VC must be
established.
IP Switching requires a protocol to
distribute the association of flow
and VCI label.
The task of cache lookup and
packet labeling is propagated
upstream to the edge of the
network.
26
Jump to first page
27
Jump to first page
Flow Classification summery




IP switch provides high speed
routing by low level switching of
flows.
It defines protocol to indicate these
flows.
All flows are classified.
The forwarding engine is optimized
for flow classification and for
forwarding uncached packets.
28
Jump to first page
29
Jump to first page
Forwarding by the IP Switch
IP Switch
Upstream
direction
IP Switch
IP Switch
Controller
IP Switch
Controller
Source
Default
Downstream
direction
Destination
Default
Default
30
Jump to first page
Forwarding by the IP Switch
Upstream
direction
IP Switch Controller
Downstream
direction
Port C
IFMP Redirect
(Flow ID,VPI/VCI=A,lifetime)
Default
z
Default
VPI/VCI=A
Port 1
Port 0
31
Jump to first page
Forwarding by the IP Switch
Upstream
direction
Downstream
direction
IP Switch Controller
Port C
IFMP Redirect
(Flow ID,VPI/VCI=A,lifetime)
Default
IFMP Redirect
(Flow ID,VPI/VCI=B,lifetime)
Default
VPI/VCI=A
VPI/VCI=B
Port 1
Port 0
32
Jump to first page
General Switch Management
Protocol (GSMP)

Simple master-slave protocol.
 Switch
controller - master.
 ATM switch - slave.


Unreliable massage transport is
assumed between controller and
switch for speed and simplicity.
GSMP runs on a single well known
virtual channel (VPI 0,VCI 15).
33
Jump to first page
GSMP (1)

The most frequent messages
(connection management) are
small enough to be a single cell.
Ver
Type
Result
Code
Transaction Identifier
GSMP Message Body
Pad (0-47 octets)
GSMP Message Format
34
Jump to first page
GSMP (2)

An adjacency protocol is used to:
 Synchronize
state across the
control link.
 Discover the identity of the entity of
the far end of the link.
 Detect when the far end is
changed.

No GSMP massages may be sent
across the link until adjacency has
been established.
35
Jump to first page
GSMP (3)

GSMP has five type of massages:
 Configuration.
 Connection
management.
 Port management.
 Statistics.
 Events.
36
Jump to first page
Ipsilon Flow Management
Protocol (IFMP)



Runs on a point to point link
between two IP switches.
The purpose of IFMP is to inform
the transmitting end of a link of the
VCI that should be associated with
a particular IP flow.
The VCI is selected by the
receiving end of the link.
37
Jump to first page
IFMP (1)

Two flow types has been defined:
 port-pair
flow (type 1) - source IP
address,destination IP address,source
port number,destination port number.
 Host-pair
flow (type 2) - source IP
address,destination IP address.

An IFMP redirect message is sent
upstream to inform the transmitter
of the association between flow
and VCI.
38
Jump to first page
IFMP (2)
Flow Type
Flow ID
Lifetime
Label
Flow Identifier
IFMP Redirect MSG
Ver
IHL
TOS
TTL
Protocol
Source IP Address
Destination IP Address
Source Port
Destination Port
Flow type 1 - Identifier
Ver
IHL
Rsrvd
TTL
Rsrvd
Source IP Address
Destination IP Address
39
Flow type 2 - identifier
Jump to first page
IFMP (3)




40
A lifetime field specified the length
of time for witch this association of
flow and VCI is valid.
The flow redirection must be
refreshed.
Flow labeling process occurs
independently an concurrently on
each link.
The flow classification policy is
consistent within an administrative
domain.
Jump to first page
IFMP (4)



When upstream and downstream
links are both labeled for a given
flow, that flow is switched directly
trough the ATM switch.
When an IP switch accepts a
redirection messages it also
change the encapsulation.
It allows an IP switch to act as a
simple based firewall.
41
Jump to first page
Conclusion






The IP switch is an alternative
architecture to the gigabit router.
It uses low level switching of flows.
It include a cooperative protocols.
Link by link basis decision.
All flows are classified.
It allows to support multicast, QoS,
Simple firewall filtering.
42
Jump to first page