munchdrabΔίκτυα και Επικοινωνίες

30 Οκτ 2013 (πριν από 4 χρόνια και 8 μήνες)

91 εμφανίσεις


Single point of failure

Hierarchical design produces many single
points of failure

Redundancy provides alternate paths, but
may undermine stability

Fig 3
1 Retana

Issues and strategies of

Redundant paths should be used only when
the normal path is broken

Load sharing must be carefully engineered
to prevent instability

Traffic shouldn’t pass through devices and
links not designed for through traffic

Redundant core design

Reduce hop count

Increase the number of failures the core can

Ring core design

Two paths for each pair of routers

Diameter is N/2

Losing a link increases the diameter

Losing any two links isolates a piece of network


Reduced number of available paths

Number of hops increases dramatically on a sing
link failure
> session timeout

Ring core

Redundant fiber ring

SONET: a pair of fiber optic links of OC
48(2.4G), 2nd fiber is used as backup

FDDI: two rings passing data in opposite
directions. If fiber fails, the ring will wrap
using dual attached nodes, healing the break

Redundancy in layer 2

Full mesh core design

Large number of alternate paths

Two hop path to any destination

N hops maximum in the worst case

Highest redundancy


Smallest hop count

Maximum redundancy

Too many alternate paths
> larger convergence
time, difficult to engineer traffic

Full mesh core

Partial mesh core design

Good compromise in hop count, redundancy and
number of paths

No more than three hops for Fig. 3

If any single link fails, maximum number of hops
will increase to four

Provides full connectivity with three links down as
long as no single router loses both of its

Some routing protocols don’t handle multipoint
partial mesh

Partial mesh core

OSPF on NBMA networks

PVCs on Frame relay, ATM has non
broadcast multiple access(NBMA)


Problem: non
broadcast, since a packet on a
subnet might not be seen by all routers connected
to it, making it hard to choose a DR(designated


1) Configure priorities to select Router A as DR

2) Treat all as point
point link
> subnet id space

3) make all the links in the multipoint network to
have same netid
> should add a lot of

to each hosts’ routing table

Distribution redundancy

Dual homing

Backup links to other distribution layer

Consider unexpected traffic patterns

Dual homing to the core

Doubles the number of available paths to
core routers, slowing network convergence

Dual homing to the core

Double avail. paths:
> Advertise only over
one link unless that link becomes unusable

Distribution routers may be drawn into core
role, passing through traffic:
> prevent D
from advertising routes learned from C back
to B

Redundant links to other
distribution layer devices


Double the core’s routing table size

Possible use of redundant paths for core

Prefer redundant links to passing core

Routing info will leak from A and spread to
other distribution layer routers, causing

Access redundancy

Dual homed
> size of routing table in the
core increases: prevent advertising
redundant links

Access redundancy

Provide redundant links in access layer:
saves one link and reduces number of
available paths
> should provide enough
bandwidth to handle traffic from both
remote sites to core

Access redundancy

Redundancy through same
distribution layer branch

Load sharing and redundancy

Redundancy through same
distribution layer branch

Filter in router G, H prevents advertising
networks not below them in hierarchy such
as D, F to C and E

Or use dial

Connections to common services

Case: what’s the best route?

How to determine best routes from a mix of
different routing metrics?

Ex) RIP uses hop count, EIGRP uses bandwidth and

Longest match > adminstrative distance > metric

Each routing protocol has a default adminstrative
distance (lower prefered)

connected: 0 static: 1, OSPF: 110, RIP: 120

floating static routes: a static route with high
adminstrative distance, 200 or more used for
backing up or conditionally advertised (BGP)

Case: Redundancy at layer 2
using switches

(Using spanning tree avoiding loops at layer
2, block ports)

up backup with a single

Configure ISDN link as a backup interface