Survey A of Light-Weight Transport Protocols for High-Speed Networks
Presented by: Satnam Singh Andrew Dzhigo
OUTLINE
Introduction to eight transport-layer protocols Define the functions under consideration Summarizes the characteristics Conclusion of eight protocols examined Future research will be devoted to demonstrating the efficacy of these options
KEY POINTS Study mechanisms
Provide: reliable, es-es transmission Examine the mechanism to support these
Conclusion
Satnam Sec. I
II III
Andrew
IV V
1
Throughput:
INTRODUCTION
Once upon a time long, long ago: (not quite, but prior to introduction of LAN/MAN)
throughput is the amount of data moved successfully from one place to another in a given time period.
Bandwidth was expensive and scarce (how did they live) Transmission errors were sufficiently frequent The throughput available to application programs was usually limited by the network. Hence, efficient error recovery processing was required.
Then came LAN and MAN
Bandwidth no longer the bottleneck
Processing: In the context of that needed to execute communication protocols
Processing consuming huge amount of resources.
Problem becomes more acute in the case of networks offering greater bandwidths e.g. FDDI (100 Mb/s) and Broadband.
Do we need the same as level of Processing as before ?
2
The Processing Redemption
Error rates have decreased significantly Buffer Overflow is the dominant source of errors
Thus, the design issues are different than before. And decisions need to be made taking in account:
Liberal use of processing power to reduce transmission cost Addition of extra processing power for error-recovery Use of simpler flow-control mechanisms
From Here to ......
Minimize processing requirements
Optimize protocol processing (for error-free class of communication)
Include more effective flow-control algorithms
3
New Applications on the Net
New technologies spur up new applications which are Bandwidth hungry.
High Resolution graphics
Computer Visualization
Interactive display of numerically intensive computation
Light-Weight protocols
These requirements necessitate the development of light-weight protocols.
What we wish to achieve with it :
u
Shorten the length of the instruction path for data transmission and reception.
u u u
Minimize the overhead for communication Also implements flow-control algorithms to prevent buffer overflow Facilitate light-weight connection management
4
Fast, High-Throughput Performance
Two approaches:
u u
Implement existing protocols such that the processing time is minimized design new protocols with mechanisms specifically architected for high bandwidth, low-delay, low-error rate communications.
Key issues: Reduction of data movement inside processing nodes the use of suitable internal queue and buffer management strategies a low overhead process structure proper embedding of the protocol in the operating system
Goal of the Paper
TCP XTP Datakit VMTP
Examine architectural mechanisms
Techniques
well suited to
High-performance communication
5
Transport Protocol Survey
Study the techniques used in LW protocols developed, and Improved implementations of existing protocols
Protocols chosen :
APPN OSI/TP4 Datakit TCP Delta-t VMTP NETBLT XTP
Survey
Purpose of the protocol
Intended operating environment
Key innovations
6
APPN
Advanced Peer-to-Peer Networking Part of IBM AS/400 and S/36 system s Built on SNA Does not define an explicit transport-layer protocol. Rather they are performed as a part of APPN. Builds on VC network service Assumes fully reliable data-link transmission *
E.g.: Setup and release functions
Datakit
Builds on VC oriented network Delivers data error-free, in sequence but possibly with gaps Intended to provide a universal data transport service Key functions include:
u u
End-to-End flow control Detection and retransmission of lost data
Universal Receiver Protocol Byte is the fundamental unit of transmission:
u
9-bit byte is used
7
Datakit
Control bytes used to request acknowledgment Lower layers relay the data in small frames (cells) Receiver must specify the loss of cells when requested so by the sender.
Delta-t
Transport interface offers a :
4 4 4
reliable end-to-end stream-oriented service
Does not guarantee delivery or duplicate protection Innovation : Development of timer-based connection management.
Supports LW connections
8
XTP
Ease of VLSI implementation Provides real-time datagrams, multicasting and efficient bulk data transfer Network + Transport = Transfer Layer Offers rate control, selective retransmission and implicit connection establishment Innovation: 4 Support for multiple addressing schemes (TCP or OSI) 4 Multicast capability 4 Regulation of Network flow as per end-user and network limitations
Transport Protocol Functions
TSDU s TSDU s
TPDU s
Transport
(packets)
Transport
Transport Layer Terminology
9
Transport Protocol Functions
u
Connection Management
u
Acknowledgments
u
Flow Control
Headings for Discussion of aspects of protocol functions
Connection Management
u
Signaling (in-band / out-band)
u
Connection Setup and Release (handshake/implicit)
u
Selecting Transport Services (Negotiation/Update)
u
Multiplexing (Yes/No)
10
TPF =>Connection Management
Signaling
Exchange of information for managing a connection Used to setup and release connections Also, managing Communication System parameters
Control information; Data In-Band Control information Data Out-Band
TCP, NETBLT, XTP, OSI/TP4, Delta-t, VMTP HIGH-SPEED ENVIRONMENT DESIRABILITY
Datakit, APPN
TPF =>Connection Management
Connection Setup and Release
Handshake: Explicit exchange of messages
Implicit or Timer-Based: Open with initial packet, close under timer control
11
TPF =>Connection Management
Connection Setup and Release
R S
Open
us ? Dup/ Previo
S
Open
us ? Dup/ Previo
R
S
R Packet received;
Connection established
ACK
Timer
Connection closed
3-way handshake
TCP, OSI/TP4 (Setup) XTP (release)
2-way handshake
NETBLT, Datakit, APPN OSI/TP4 (release)
Implicit
Delta-t, XTP (Setup)
Acknowledgements
Sender-dependent based on reception data: TCP, VMTP, NETBLT, Delta-t, OSI/TP4 explicitly solicited by sender: Datakit, XTP Sender-independent: OSI/TP4 (by option) required at the transport layer: APPN Not
12
Flow and congestion control
End-to-end flow control: Datakit, NETBLT, Delta-t, VMTP, OSI/TP4, XTP, TCP Access control: explicit: XTP implicit: VMTP, TCP required at the transport layer: APPN Not
Flow and congestion control (continued)
Window Flow Control: cumulative window: NETBLT abso...