ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
1
ATM-Sat
Intersatellite Link (ISL) Networks: Topological Design, Routing and
Network Dimensioning
Markus Werner
DLR Oberpfaffenhofen
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
2
ATM-Sat
4 Why (still) ISLs ? ... and what about ISLs in ATM-Sat ?
4 Which ISL network topology ?
h from the Iridium heritage to a modern (sustainable) design !
4 “ATM-based“ ISL routing concept:
h a discrete-time dynamic routing framework derived from core ATM principles rather than an ATM implementation !
4 Integrated ISL routing and network dimensioning
4 Routing towards an ISL future:
h ATM+IP ? ↔ all IP ? ↔ MPLS ↔ optical networks . . .
4 A conclusion: Good concepts are sustainable - sustainable concepts are good!
Outline
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
3
ATM-Sat
4 Iridium&ISLs: final desaster or lessons learnt (to learn) ?
4 Teledesic: ongoing dream or diminishing nightmare ?
4 Backbone-in-the-sky (not only “Internet-in-the-sky“) as a sustainable concept
4 A sustainable ISL networking concept will pay off one (resurrection) day !
Why (still) ISLs ?
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
4
ATM-Sat
... and what about ISLs in ATM-Sat ?
4 We don‘t have (short-term) stock options in ISL networks
4 No “early-to-market“ strategy, but strategic project ...
4 ... consequently re-defining or just recalling the orientation/direction:
h “friendly“ scenarios: – high-capacity multicast– fixed terminals – aggregated traffic ...
h prospective markets: – high quality Internet– particular global VPNs– high-speed and reliable global information distribution– trunking market niches ...
h driving technologies: – MPLS– optical networking– λ switching ...
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
5
ATM-Sat
Scenario
Space segment:ISL trunk network
Earth segment: Wireline backbones
Air interface:OD traffic
ISLSubnetwork
ATM, IPUp/Downlink:- Handover- Sat. Diversity
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
6
ATM-Sat
Problems with (Near-)Polar Orbits: The Iridium HeritageSatellite constellation:
"Seam" betweencounter-rotating orbits:
Snapshot of footprints and ISL topology:
intra-plane ---- inter-plane ISLs
Drawback: two kinds of irregularityno ISLs crossing the "seam"deactivation of inter-plane ISLs in polar regions-> considerable path rerouting requirements !
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
7
ATM-Sat
The Inclined Walker “Delta“ Constellation M-StarConstellation:
regular phasing, phasing factor 5absolute symmetry of the orbital constellationno "seam" like in polar constellations
orbit altitude 1350 kmorbit period 113 min# of satellites 72# of orbits 12inclination 47°
Planar projection (schematic):
M-Star utilizes the promising combination of “delta“ constellation pattern and optical ISLs
Equator
Equator
Equator
North Pole
South Pole
1
6
37
38
39
40
41
36
30
35
34
33
32
31
26
27
28
29
24
2519
18
23
22
21
20
15
16
17
12
13
1436
41
40
39
38
37
43
44
45
46
47
42
48
49
50
51
52
53
54
55
56
57
58
59
64
63
62
61
60
6571
66
67
68
69
703
8
9
10
11
2
3
4
5
7
0
Orbits
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
8
ATM-Sat
Reference ISL Topology for M-Star
Snapshot at t=0:
All links are permanently maintained over the whole orbit period !
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
9
ATM-Sat
Network Design: Concepts and Methods
4 Network design = topology + routing + dimensioning
h usually an iterative process
4 ISL network design:
h extreme challenges: dynamic topology, traffic variation
h design process must be simple !
4 Ingredients for “simple“ ISL network design:
h hierarchical “open-loop“ design:begin {topology; routing; dimensioning} end
h top-down decomposition
h standard modules (shortest path search, LP optimization, ...)
h abstraction of dynamics: discretization and virtualization
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
10
ATM-Sat
Discrete-Time Dynamic Routing FrameworkOff-line Routing Framework
t=T-∆t s=S-1
t=∆t s=1
t=0 s=0Instantaneous
Virtual Topology Setup I-VTS(s)
t=2∆t s=2
Discrete-Time VirtualTopology Design DT-VTD
(1,2)
(1,3)
(N-1,N)
K Shortest Path Search
KSPA
OD pairs∀
ProcessingOD)(kK →
flexible k , path grouping / path reordering
Topological DesignPermanent Physical Topology Design P-PTD
Topological design of ISL networkwith permanent (non-switched) physical links
Topological design of ISL networkwith permanent (non-switched) physical links
for Walker delta constellations
Capacity DimensioningDiscrete-Time Multi-Step Dimensioning DT-MSD
Single-stepdimensioning
+ post-processing
Linkedsingle-step
dimensioning
Multi-stepdimensioning:
Dynamic optimiz.
or or
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
11
ATM-Sat
Path Grouping for Origin-Destination (OD) Pairs
equator
e
north pole
38
39
40
30
35
34
24
2536
41
40
39
38 44
45
46
47
4 2
49
5051
5253
5 5
56
57
58
59
62
6166 3327
2 8
29
19
18
23
14
64
63
67
69
70 3
10
11
2216
17
12
1387
6
2
68
5
1
0
group1
group2
group2
group2
group2
27
1
23
64
3
67
69
70
1122
16
17
12
13
8
7
6
2
68
5
1
0
group1
group2
group2
group2
group2
Hop-based path groupingfor example OD pair:
Clear path group separation by cost ranges(costs = accumulated propag. + processing delays)
PG separation extends over all steps; no overlaps of cost ranges!Ordering of paths (KSPA) can only vary within a group
Select k* = k(OD) such that k*-path set always forms a complete path group
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
12
ATM-Sat
Network Dimensioning: Target Functions
ISL capacity requirements
Bandwidth & RF power
Satellite capacity requirements
Processing power & buffer sizes
DC power
Size & weight of satellite System costs
positive “side effect“: better utilization of installed network capacity
Candidate target functions: TF1: Minimize worst case link (WCL) load
TF2: Minimize worst case node (WCN) load
LEO constellation dynamics --> every sat/link encounters worst case sometime--> all sats/links to be dimensioned accordingly
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
13
ATM-Sat
Network Dimensioning: Approaches
ES: Equal sharing of total OD traffic between all k OD pathsFO: Linear optimization of OD traffic distribution on k alternative paths
without specific constraintsBO: Linear optimization of OD traffic distribution on k alternative paths
with additional constraints
Discrete-Time Multi-Step Dimensioning DT-MSD
Single-stepdimensioning
+ post-processing
Linkedsingle-step
dimensioning
Multi-stepdimensioning:
Dynamic optimiz.
or or
isolated step history-based
Equal Sharing (ES) iES hES
Bounded Optimization (BO) iBO hBO
Full Optimization (FO) iFO hFO
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
14
ATM-Sat
LP Optimization Approach: Isolated StepFor each time step s = 1 ... S :
Post-processing : Take maximum of all S minimax values
for k=3:
7668optimization
variables
144constraints
2556constraints
7668constraints
iFO
iBO
)}({max pnlln
∀
subject to
wnnwPpp
pw ∀= ∑∈
,,
minimizeminimax problem
lnnnwPpp
ppl
Wwwl ∀=≥ ∑∑
∈∈
,,,
max δ
)(max pnn
subject to
minimizelinear optimization problem
wnnwPpp
pw ∀= ∑∈
,, OD/path capacity constr.
minimize WCL capacity
link capacity constraint
pnn wp ∀∈≤≤ ,]10[,0 Kααpath cap. upper bounds
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
15
ATM-Sat
Performance of Isolated Step Optimization from here, all numerical examples assume network-uniform traffic,i.e., a normalized symmetrical traffic load of 1 between all OD pairs
Worst Case Link (WCL) Load: iES - iBO - iFO
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
16
ATM-Sat
History-Based Optimization: Rationale
4 Modeling deficiencies of isolated step approach:
h assumes uncorrelated demand pair capacities
h neglects “history“ of single calls
h implicitly assumes that all calls are freely (re)routable in each step
⇒ isolated step optimization results are “too good“ (considering QoS)
4 History-based approach:
h guarantees that remaining (old) calls stick to the once chosen path:
→ avoid uncontrolled delay offset→ reduce dropping probability→ avoid unnecessary signaling
h takes care of incoming rerouted calls (from other OD pairs; sat. handover!)
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
17
ATM-Sat
History-Based Optimization: Modeling
4 Call model:
h determines the ratio between old and new traffic
h classical Erlang model: Poisson arrival, negative exponential holding time
4 Rerouting model:
h sub-classifies the old traffic into remaining and rerouted
h based on a handover model for serving satellites (source/destination)
With ODT(s) = OD demand pair Traffic at step s:
total ODT(s) = new ODT(s) + rerouted ODT(s) + remaining ODT(s)
old ODT(s)freely (re)routable
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
18
ATM-Sat
From Isolated Step to History-Based Optimization
lnnnwPpp
ppl
Wwwl ∀=≥ ∑∑
∈∈
,,,
max δ
)(max pnn
subject to
minimizeisolated step approach
wnnwPpp
pw ∀= ∑∈
,, OD/path capacity constr.
minimize WCL capacity
link capacity constraintFO
BO pknn wp ∀∈≤≤ ,]1/1[,0 Kααupper path cap. bounds
history-based approach
lower path capacity bounds
K,remwpp nnn α≤≤
lower and upper path cap. bounds
pnn pp ∀≤ ,rem
unchanged
plus:
in other words:rernewremopt 1 pppp nnnn +=−=
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
19
ATM-Sat
WCL Load: Isolated Step vs History-Based
number of alternative OD paths k
0 1 2 3 4 5 6 7 8
WC
L tra
ffic
load
(mea
n)
0
100
120
140
160
180
200
220
240hFOiFOhESiES
(mean call holding time: 3 min)
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
20
ATM-Sat
Link Load over Time
4 periodic behaviour: peaks at higher latitudes (shorter inter-orbit links)
4 inter-orbit ISLs are the critical ones
north south
equator equator
for shortest path routing (no optimization):
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
21
ATM-Sat
Link Load Distribution in the Network: hFO vs iFO
PL identification number0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
PL tr
affic
load
0
40
50
60
70
80
90
100
110
120
iFOhFO
Snapshot at t=0:
inter-orbit
intra-orbit
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
22
ATM-Sat
Link Load Distribution in the Network: hFO vs iFO
Snapshot at t=0:
PL number (descending load value sorted)0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
PL tr
affic
load
0
40
50
60
70
80
90
100
110
120
hFO inter-orbithFO intra-orbit
iFO inter-orbitiFO intra-orbit
inter-orbit
intra-orbit
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
23
ATM-Sat
... and all that on the map:without optimization (shortest path routing):
with optimization (multipath/alternate routing):
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
24
ATM-Sat
Conclusions (on Work Done)
Network design process for LEO ISL networks:
Topological design and dynamic routing framework
Network dimensioning: (a) heuristic design rules, (b) LP optimization approach
Advanced history-based approach shows limited reduction of optimization gain
Simple isolated step approach remains a low-complexity method for “quick-and-dirty“ dimensioning
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
25
ATM-Sat
A Motivation for Ongoing Work:Connection-Oriented (CO) vs Connectionless (CL) ?Looking from a CO engineer‘s point of view:
• ISL topologies are (deterministically) dynamic• end user connections must be handed over between serving satellites
==> “connections“ must be broken up and rerouted in predictable steps
Looking from a CL engineer‘s point of view:
• ISL topologies are dynamic (-> routing updates ?!); YES, BUT:• ISL dynamics is predictable, periodic• ISL topologies are regular (in terms of nodes and links)
==> use nice properties, become “a bit more CO“?
Looking from an unbiased engineer‘s point of view:
==> Combine benefits, be SCO (sub-connection oriented)!
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
26
ATM-Sat
Packet-Oriented ISL Networking4 Ongoing joint research area of Institut Jožef Stefan (IJS) Ljubljana and DLR
h partly within COST 252 Action
h ISL network simulator on packet level implemented at IJS:– Dijskstra SPA for route calculation– traffic adaptive components using queue status information– generic on-board packet generation approach– interface to up/donwlink modules and corresponding traffic input
4 Some results:
h constraint-based routing combines topology and load information
h adaptive routing concept and protocol implementation verified
h some performance studies (routing update invterval, WCL load, ...)
h alternate link routing proposed for “optimized“ flows ...
4 ... but still pure IP ... no systematic traffic engineering ... -> IP over ATM??
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
27
ATM-Sat
Multi-Protocol Label Switching (MPLS)4 THE solution for future packet-switched backbones (?)
4 THE way to integrate ATM and IP (?), combining the benefits
h calculate CO paths (ATM) using routing protocols like OSPF (IP)
h establish these Label Switched Paths (LSPs) using MPLS unique protocols
h allow frame and cell forwarding on the same link & in the same router
4 Appeal of and issues for MPLS-based ISL networking
hMPLS is a technique dedicated to backbones
h a “blue-sky“ ISL network may be an ideal basis to exploit the fully potential of MPLS in a homogeneous global MPLS domain
h each satellite acts as edge router (serving ground) and core router (transit)
h importance of traffic engineering; potential use of proposed dimensioning methods, alternate routing approaches ...
h use OSPF and LDP protocols within the proposed routing framework ?!
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
28
ATM-Sat
Optical ISL Networking4 Key role of optical communications for current/future backbone networks
h (D)WDM - (dense) wavelength division multiplex
h wavelength-routing or λ-switching
h form optical transport network (OTN) by circuit-switched lightpaths
4 For future ISL backbones:
h low mass, size and power consumption of optical ISL terminals
h PAT requirements for inter-plane ISLs can be met by laser technology
h space OTN concepts are really close to the developed connection-oriented routing and dimensioning framework
4 Research on space OTN has been initiated
h lightpath assignment, (even more) simplified routing, new dimensioning issues, interoperation between optical and electronic domain (esp. for efficient resource utilization, flexible traffic flow multiplexing, ...)
ATM-Sat Workshop, 12 December 2000, Oberpfaffenhofen KN-DNMar
kus
Wer
ner
29
ATM-Sat
Another Conclusion
Good concepts are sustainable –
sustainable concepts are good
(and both are required)
Just do it !