ATM Traffic Management in a LMDS Wireless Access Network

1
ATM Traffic Management in a LMDS Wireless Access
Network

• Josué KURI, Maurice GAGNAIRE
• Ecole Nationale Supérieure des Télécommunications
• 46 rue Barrault - 75634 Paris CEDEX 13, FRANCE
• Tel [phone] Fax [phone]
• E-mail kurigagnaire_at_enst.fr
  http//www.enst.fr/kuri

2
Outline

• Introduction the LMDS system at a glance.
• MAC-PDU format.
• LMDS-DAVIC frames.
• Unspecified points in the DAVIC standard and
  proposals.
• VBR and GFR traffic source models.
• Node configuration.
• Simulation results.
• Conclusions.

3
Introduction

• European context
• Copper wire unbundling limited to Germany and
  Sweden.
• National regulation bodies open the WITL market
  by the end of 2000.
• Two variants of LMDS in Europe
• 3.5 GHz (500 MHz band) Residential users.
• 26 GHz (1 GHz band) Professional users.
• Standarization bodies
• LMDS-DAVIC.
• LMDS-ETSI.

4
Introduction
Video Server
Micro cell
NIU
AIU
Fiber feeder network
NIU
Local Exchange
NIU
Micro cell
Micro cell
5
Introduction
Downstream and upstream spectrum allocation
Downstream
DQPSK modulated channels
QPSK/QAM modulated channels
2 MHz
40 MHz
Upstream
6
MAC PDU Format

• Downstream (1)
• MPEG2 Transport Stream (digital video).

MPEG2-TS downstream flow
204 bytes
204 bytes
204 bytes
7
MAC PDU Format

• Downstream (2)
• ATM (data and signaling).

Packet n
ATM transport MUX packets (two-packet sequence)
Packet n1
7 ATM cells 3 control bytes per two-packet
sequence
Packet n
Packet n1
8
MAC PDU Format

• Upstream
• ATM (data and signaling).

5 bytes
48 bytes
4 bytes
53 bytes
10 bytes
1 byte
Preamble
RS(63,53)
Guard
68 bytes
ATM upstream flow
9
LMDS-DAVIC Frames

• Downstream/Upstream frames are 3 to 6 ms.
• Downstream frame.
• 728 slots (frame of 5.819 ms).
• FS frame start, NIU random access.

10
LMDS-DAVIC Frames

• Upstream frame
• 24 slots (frame of 5.819 ms).
• C contention, P polling, NIU reserved.
• Remark No piggy-backing!

11
Unspecified Points inDAVIC and Proposals

• Contention Resolution Algorithm.
• Contention slot C allocation strategy (AIU).
• Bandwidth allocation strategy (AIU).
• Bandwidth request strategy (NIU).

12
Proposal 1 ContentionResolution Algorithm

• Immarsat CRA

• Activated in the NIU if a C slot collides.
• Set R?0.
• Randomly select n?0..3R.
• Wait n upstream frames before slot
  retransmission.
• If retransmission collides, R?R1 (up to 3) and
  goto 2.

13
Proposal 2 ContentionSlot C Allocation

• Alternatives
• Static.
• Dynamic, according to observed collisions.
• Proportional to the number of active NIUs.
• Our proposal
• 15 of the number of active NIUs.

14
Proposal 3 Bandwidth Allocation Strategy

• The AIU must know the ATM traffic contract to
  serve bandwidth reservation requests accordingly.
• A request of a given connection is sent by the
  NIU (in a C slot) at the beginning of a burst.
• Slots are reserved to the NIU until a
  non-utilized slot reserved slot is received by
  the AIU.

15
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

16
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
17
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
18
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
19
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
20
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
21
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
22
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
23
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
24
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
25
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
26
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

Toward AIU
NIU queue
27
Proposal 3 Bandwidth Allocation Strategy

• Bursty traffic increases mean access delay.
• Burst merging Cells of a new burst can use the
  bandwidth allocated to the previous burst if it
  has not sent a non-utilized slot to the AIU.

NIU queue
28
VBR source model

• VBR connections modeled with on-off sources.
• Traffic descriptors
• Peak Cell Rate (PCR)
• Sustainable Cell Rate (SCR)
• Burstiness

rate
PCR
SCR
time
Ton
Toff
Ton
Toff
Ton
29
VBR source model

• Ton exp(?)
• Tcell Transmission time of an ATM cell at PCR.
• Worst case is considered by imposing a
  transmission at SCR between successive bursts
• Thus, the value of Toff is deterministically set

30
GFR source model

• Alternatives for IP over ATM RFC 1483, RFC 1577,
  LANE.
• We use RFC1483 with the Guaranteed Frame Rate
  transfer mode.
• GFR guarantees
• A Minimum Cell Rate (MCR) to connections.
• Fair share of uncommited bandwidth among
  connections.

31
GFR source model

• RFC 1483 encapsulation.

• IP message length distribution.
• Packet inter-arrival is exponential.

TCP
UDP
IP
LLC/SNAP
AAL5
ATM
32
Node Configuration

• NIU. Per-VC queueing and W Round Robin scheduling.

AAL2
WRR
VC1
VC2
VBR
VCn
LLCSNAP/AAL5
DT EPD
VC1
VC2
IP
VCm
33
Node Configuration

• AIU. Bandwidth allocation on each frame to NIUi
• Active VBR connections (1..n)
• Active GFR connections (1..m)
• Total bandwidth

34
Node Configuration

• AIU. Allocation of uncommited (extra) bandwidth
  among active GFR NIUs.

35
Simulation results

• Ideal physical channel.
• A single upstream channel is considered.
• NIUs are all at the same distance of the AIU.

36
Simulation results

• CRA performance under Poisson traffic in C
  slots.

600
500
400
Throughput (Kbit/s)
300
200
100
0
Offered load
37
Simulation results

• Performance under voice traffic.
• VBR sources with ITU-T G.726/G.727 ADPCM rate
  compliance
• A strict two-way end-to-end delay bound of
  ?15-30 ms is required for voice calls.

With AAL2/ATM overhead
38
Simulation results

• Performance under voice traffic.
• Mean cell access delay vs active NIUs.
• 1 VBR connection per NIU.

Mean cell access delay (ms)
NIUs
39
Simulation results

• Performance under TCP/IP traffic.
• A TCP/IP source generates R80 Kbit/s, in
  average. (5 of upstream channel capacity 1.583
  Mbit/s).
• Two types of MCR allocated to the connection

40
Simulation results

• Performance under TCP/IP traffic.
• Aggregated goodput vs active NIUs.
• 1 TCP/IP connection per NIU.

Aggregated goodput (Kbit/s)
NIUs
41
Simulation results

• Performance under TCP/IP traffic.
• Mean cell access delay vs active NIUs.
• 1 TCP/IP connection per NIU.

Mean cell access delay (ms)
NIUs
42
Simulation results

• Performance under multiplexed traffic.
• NNIU 8 (Number of NIUs).
• Ncnx (Number of
  connections in NIUs).
• ? 0.2, 0.4, 0.8 (Offered load).
• C 1.583 Mbit/s (Upstream channel capacity).
• Reserved bandwidth

43
Simulation results

• Performance under multiplexed VBR traffic.
• Mean cell access delay vs Ncnx.

350
300
250
200
Mean cell access delay (ms)
150
100
50
0
2
3
4
5
6
7
8
9
1
Connections/NIU
44
Simulation results

• Performance under multiplexed TCP/IP traffic.
• Mean cell access delay vs Ncnx.

600
550
500
450
400
Mean cell access delay (ms)
350
300
250
200
150
100
50
2
3
4
5
6
1
Connections/NIU
45
Conclusions

• In a LMDS cell with 205 upstream channels, around
  10000 VBR connections (voice calls) are
  achievable.
• Extra bandwidth is efficiently allocated among
  TCP/IP connections.
• Burstiness of TCP/IP connections induces high
  access delay.
• It is preferable to distribute multiplexed VBR
  connections over different upstream channels.

46
Conclusions

• We will investigate mux of VBR anf TCP/IP
  connections in the NIU.
• More sophisticated service disciplines must be
  implemented to satisfy VBR and GFR contract
  rates.