PPT Version

draft-dachille-diverse-inter-region-path-setup01.txt
A. D’Achille [email protected]
U. Monaco [email protected]
V. Sharma [email protected]
F. Ricciato [email protected]
D. Ali [email protected]
M. Listanti [email protected]
61th IETF, Washington DC
November 2004
Joint Selection Approach Overview (1/3)
• Addresses problem of implementing end-to-end diverse routing
across multi-region networks
• Flexible network model -- applies to any network where
– Network topology can be modeled as a conglomerate of
interconnected regions (OSPF areas, ASs, etc.)
Head
BN1
BN3
Region 1
BN2
BN5
BN7
Region 2
BN4
Region 3
BN6
Border Node
BN8
Tail
working path
backup path
– Border nodes have:
 Complete knowledge of intra-region topology
 Summarized info. of inter-region connectivity (list of interregion paths)
61th IETF, Washington DC
November 2004
Joint Selection Approach Overview (2/3)
•
New RSVP-TE object, Associated Route Object (ARO), to support the
simultaneous computation of the disjoint LSPs
Records paths of disjoint LSPs during the signaling of the first LSP
The format of the sub-objects included in ARO is shown below
•
•
Tail-end
Head-end
Tail-end
1
Primary path
Secondary path
a)
PAT
H
poss (ERO E
ible
x
ARO pansion
,
Expa
nsio
n)
ARO
ERO
V(
RES
b)
H
Type
16
Area_ID
Type
16
8
1
Type
32
IPv4 address
Lenght
Addr_len
IPv4 address (continued)
c)
32
Lenght
8
1
)
AR O
)
(ERO
PAT
8 9
Resvd
16
32
IPv6 address
Lenght
…
RES
V
t
IPv6 address (continued)
Addr_len
t
t
RSVP-TE signaling for JSA
61th IETF, Washington DC
November 2004
ARO Sub-objects
Resvd
Joint Selection Approach Overview (3/3)
Advantages of the disjoint path computation scheme are:
• Independence from disjoint paths computation algo./method
– E.g. Bhandari, Surballe, others, …
• No extra signaling/routing load to achieve the objective
• Generic method of communicating alternate path info.
– Usable in other contexts
• Robustness against trapping topology problem
• Better performance on path selection
61th IETF, Washington DC
November 2004
Experimental Results (1/4)
•
Comparison of different inter-region routing mechanisms:
1) JSA (ARO)
•
2) ISPA (RRO+XRO)
3) Global optimum
Simulation scenario:
–
24 inter-region networks of 3 regions connected linearly
–
Intra-region topology derived from realistic router-level ISP
topologies (abovenet, ebobe, tiscali, exodus … more than 200
routers each) [from http://kom.tu-darmstad.de/˜heckmann]
–
Two border node selection methods
•
•
–
Random selection
Max degree selection
500 instances of inter-region double path computation
demands
61th IETF, Washington DC
November 2004
Experimental Results (2/4)
2 BN
3 BN
BN random selection
High ISPA trapping
61th IETF, Washington DC
November 2004
Experimental Results (3/4)
3 BN
2 BN
BN max degree selection
61th IETF, Washington DC
November 2004
High ISPA trapping
Experimental results (4/4)
1)
JSA (ARO) almost always finds diverse paths (if they exist)
at the first shot
–
2)
ISPA (RRO+XRO) was trapped in some topologies at the first
shot (and should therefore revert to crankback)
–
3)
Failed in 2-3 cases over thousands of simulation runs
Happens approx. in 10% of cases
When diverse paths are found by all the three methods
–
–
No substantial difference in the overall cost between JSA
(ARO) and optimum (min-hop metric) [+1.5% on average]
Slight worse perf. of ISPA (RRO+XRO) [+6% on average]
Detailed summary report available at
http://www.metanoia-inc.com/Publications/Interregion-Diversepath-Perfcomp.pdf
61th IETF, Washington DC
November 2004
Relation with PCE
•
With respect to the PCE architecture [draft-ash-pcearchitecture-00.txt ]:
–
–
•
JSA (ARO) is consistent with Multiple PCE path computation
with no inter-PCE communications
Could be used in both region “Centralized/Distributed
computation models”
With respect to the PCS/PCE path computation
scheme [draft-vasseur-ccamp-inter-domain-pathcomp-00.txt ], JSA (ARO):
–
–
Achieves same performance as the optimal method "almost
surely" in practical cases (see from simulations, no formal
proof yet)
Involves less information exchange and signaling overhead
(better scalability)
61th IETF, Washington DC
November 2004
Next Actions
• Improve & extend performance analysis/simulation
• Open to suggestions from the community …
– Carrier/provider input about:
• Alternative criteria for selecting the border routers?
• Other typical network topologies to test?
– Any input about the scenario(s) for future simulations?
• Obtain further WG feedback/inputs on the JSA (ARO)
scheme
• Progress the document to a WG document
61th IETF, Washington DC
November 2004