Transcript Distributed Protocol Stacks

19-23 May 2008
Beijing, China
Distributed Protocol
Stacks
Dzmitry Kliazovich and Fabrizio Granelli
University of Trento, Italy
TCP/IP Layering

TCP/IP protocol reference model, the de facto standard in
Internet, is based on ISO/OSI layering principles


Reasons: Interoperability and Fast Deployment
TCP/IP layering drove the success of Internet in 90s

Demonstrates good performance in homogeneous networks
which rely on similar PHY technologies
However, nowadays networks are
different - Heterogeneous !!!
26 - 30 November 2007
Fabrizio Granelli ([email protected])
Heterogeneity

Network Heterogeneity
Devices: hand-held, mobile, sensors
 Links & protocols: Wireless and Wired
 Services: variety of applications, data and services
 Need for rapid deployment and organization


Leads to poor TCP/IP performance, due to


Strict layering
Lack of cooperation and coordination
C. Barakat, E. Altman, and W. Dabbous, “On TCP performance in a
heterogeneous network: a survey”, IEEE Communications Magazine,
vol. 38, no. 1, January 2000, pp. 40 – 46.
26 - 30 November 2007
Fabrizio Granelli ([email protected])
TCP/IP optimization in
Heterogeneous Networks

Cross-Layering




Allows awareness and cooperation between protocol layers
Joined, coupled design
Examples: TCP notification of non-congestion related losses
Agent-Based Networking


Aims at introduction of active functionalities in the passive
network core
Examples: Web-proxy, cashing proxy
26 - 30 November 2007
Fabrizio Granelli ([email protected])
Our Approach
Layered Design
Cross-Layering
Agent-based
Networking
Distributed Protocol Stacks

Idea: Extend the idea of protocol stack
modularity making it network-wide
26 - 30 November 2007
Fabrizio Granelli ([email protected])
Distributed Protocol Stacks

Design Details



Each functional block of the protocol stack (a protocol layer or
its part) can be abstracted into a separate module and
implemented at a different node in the network
Communication between host protocol stack and removed
module is performed using a custom “lightweight” protocol
Applications (or What is it for?)


Move protocol stack functions that generate high communication
overhead into the network core behind the bottleneck link
Caution: not all the protocol stack functions can be abstracted
and separated
26 - 30 November 2007
Fabrizio Granelli ([email protected])
Distributed Protocol Stacks
26 - 30 November 2007
Fabrizio Granelli ([email protected])
Distributed Protocol Stacks

Design Procedures

Abstraction: identification and isolation of a
particular function or set of functions of the
protocol stack

Detachment: separation of the chosen function or a set of functions into
a standalone functional block as well as its registration, transfer, and
execution with “friendly” network elements (routers, switches, or
gateways)

Communication between detached functional block and the host
protocol stack

Execution: triggered by the host node or performed automatically by
Module Running Environment (MRE) implemented by the network
element
26 - 30 November 2007
Fabrizio Granelli ([email protected])
Distributed Protocol Stacks

Concerns
Interoperability
 Incremental deployment


Benefits



Protocol stack performance
Enablers for new user applications
Driven by network operator perspective
Design of Distributed Protocol Stack solutions should be
driven by cost/benefits analysis !!!
26 - 30 November 2007
Fabrizio Granelli ([email protected])
A Case Study
3G LTE background
Bottleneck
link
IP network
File Server
Transport
User Terminal
Base station
TCP data
Transport
Output()
TCP ACKs
TCP ACKs
Network
Link
Physical
Module running
environment
Link
Physical
Network
Link
Link
Physical
Physical
Custom
protocol
26 - 30 November 2007
ACK gen
Fabrizio Granelli ([email protected])
ARQ Proxy - Approach
File Server
Evolved Packet
Core
User Equipment
(UE)
Enhanced Node B
(eNB)
ARQ Proxy
TCP
MAC
MAC
ARQ Client
TCP
TCP Data
PHY /LL
Headers
TCP
Data
TCP Data
o Access TCP header
Generate
TCP ACK
o Get IP addr, port, etc.
o Generate TCP ACK & store
TCP ACK
Index
HARQ TCP
+
ACK
Index
TCP ACK
26 - 30 November 2007
Fabrizio Granelli ([email protected])
TCP ACK
Index
o Compute TCP ACK
identification index
ARQ Proxy - Benefits

Reduced RTT (Round Trip Time)
TCP Data
IP Network
Fixed Host
(FH )
Enhanced Node B
(eNB)
Mobile Node
(MN)
TCP ACK
Medium Access + TCP ACK
Transmission over radio channel
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
ARQ Proxy - Benefits

End-to-end TCP semantics are maintained
TCP Data
IP Network
Fixed Host
(FH )
TCP ACK
Dzmitry Kliazovich
([email protected])
Mobile Node
(MN)
Enhanced Node B
(eNB)
LL- ACK
Sept. 11, 2007
ARQ Proxy - Benefits

Higher channel error rates tolerance
Wired channel
(BER = 10-6 to 10-8)
Wireless channel
(BER = 10-3 to 10-1)
IP Network
Fixed Host
(FH )
Enhanced Node B
(eNB)
Mobile Node
(MN)
TCP ACK
No TCP ACK over radio channel
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
ARQ Proxy - Benefits

Mobility & Incremental deployment
No TCP state
related information
is maintained
IP Network
Fixed Host
(FH )
Enhanced Node B
(eNB)
ARQ Proxy
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
Mobile Node
(MN)
ARQ Client
Current Activities and Future Work

ARQ proxy material
 ARQ
proxy for WiFi (IEEE 802.11)
• F. Granelli, D. Kliazovich, J. Hui, and M. Devetsikiotis, "Performance Optimization of SingleCell Voice over WiFi Communications Using Quantitative Cross-Layering Analysis," 20th
Iternational Teletraffic Congress (ITC'20), Ottawa, Canada, June 2007.
 ARQ
proxy for 3G LTE (Super-3G)
• D. Kliazovich, F. Granelli, S. Redana and N. Riato, “Cross-Layer Error Control Optimization
in 3G LTE,” IEEE Global Communications Conference (GLOBECOM), Washington, DC, U.S.A,
December 2007.
 ARQ
proxy for WiMAX (IEEE 802.16)
• Submitted for publication in Globecom 2008.
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
Current Activities and Future Work

ARQ proxy material
 EU
patent-pending
• D. Kliazovich, F. Granelli, S. Redana, and N. Riato, “Cross-Layer Error
Recovery Optimization for 3G LTE Systems,” EP 07425087.9
 Proposal
to 3GPP standardization group
 Under
consideration for next generation equipment
produced by Nokia Siemens Network (NSN)
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
Conclusions

Reallocation of protocol stack functions in the network
creates an additional degree of freedom for optimization

The proposed Distributed Protocol Stacks are considered
for Next Generation Internet (NGI)

Brings novel applications, optimal protocol stack
performance, and better configurability at the expense of
reduced interoperability and incremental deployment
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
Thank you!
26 - 30 November 2007
Fabrizio Granelli ([email protected])
ARQ Proxy – Packet Identification

3G LTE: Hash values

WiFi: Frame Sequence Numbers
PHY
LL
Header Header
Address 3
B0
4
ARQ (HARQ-ACK)
or
HARQ
ACK
Enhanced Node B
(eNB)
26 - 30 November 2007
Sequence
Address 4
Control
B15
B3 B4
Fragment
Sequence Number
Number
Bits:
Hybrid
TCP data
User Equipment
(UE)
Fabrizio Granelli ([email protected])
12
ARQ Proxy - Limitations

TCP ACKs are not substituted:
 During
connection establishment and connection
termination packets (identified by SIN and FIN flags)
 For TCP ACK encapsulated into TCP data packet in
case of bidirectional data transfer
 For Duplicate TCP ACKs
 TCP ACK advertising exhausted receive buffer
resources (rwnd field)
Dzmitry Kliazovich
([email protected])
Sept. 11, 2007
Evaluation: 3G LTE scenario

Round Trip Time (RTT)

90
0.7
80
Average TCP Throughput (Mb/s)
0.6
70
RTT (ms)
60
50
40
30
20
10
0
Hash value errors
20
Sept. 11, 2007
40
60
Simulation time (s)
0.4
0.3
0.2
0.1
TCP
TCP + ARQ Proxy
0
0.5
80
100
0
0
0.1
0.2
Dzmitry Kliazovich ([email protected])
0.3
0.4
0.5
Hash error rate
0.6
0.7
0.8
WiFi: Evaluation Results

Throughput performance
ARQ proxy ON
ARQ proxy OFF
6
Improvement: 20%
Throughput (Mb/s)
5
4
3
2
Improvement:1
up to 100%
VoIP and Multimedia
Applications
Dzmitry Kliazovich
([email protected])
0
0
300
600
900
1200
TCP/IP datagram size (Bytes)
Sept. 11, 2007
1500
TCP file transfer,
Ethernet MTU
WiFi: Evaluation Results

Round Trip Time (RTT) improvement
60
ARQ proxy OFF
Order of
milliseconds
Smooth RTT (msec)
50
40
30
ARQ proxy ON
20
10
0
Dzmitry Kliazovich
([email protected])
0
10
20
30
40
Simulation time (seconds)
Sept. 11, 2007
50
WiFi: Evaluation Results

High error rate tolerance
7
ARQ proxy ON
ARQ proxy OFF
6
Throughput (Mb/s)
5
4
3
2
1
0
Dzmitry Kliazovich
([email protected])
0
0.05
0.1
0.15
0.2
0.25
Packet Error Rate (PER)
Sept. 11, 2007
0.3
0.35