Transcript How Ethernet becomes industrial

Workshop : Ethernet as a fieldbus
Genève – Friday 28th September 2001
How Ethernet becomes industrial
Prof. Bernard Jouga - Supélec, Rennes – France
[email protected]
You said Ethernet…
Ethernet offers


A level 1 specification : cabling systems, physical
layer… in several versions
A level 2 specification : Medium Access Control…
in several versions
So many differences between vintage Ethernet
and future Ethernet !
Friday 28th
September 2001
CERN Workshop Ethernet
2
… What do you mean ?
Ethernet doesn’t offer





A complete stack up to applications, neither for the
office nor for the plant !
One standard solution for determinism and QoS
issues (not yet ?)
One standard solution for connecting in industrial
environment (not yet ?)
One standard solution for redundancy management
A satisfying solution for accurate stations
synchronization
As many Ethernet solutions as network vendors ?
Friday 28th
September 2001
CERN Workshop Ethernet
3
The everlasting confusion
"I need Ethernet if I need a network able
to support TCP/IP applications"
Friday 28th
September 2001
CERN Workshop Ethernet
4
Agenda
Determinism and Qos with Ethernet


"Vintage" Ethernet
Switched Ethernet
 Which QoS implementation ?
An overview of vendors proposals



Fieldbus interconnection with Ethernet/TCP/IP
Fieldbus protocols over Ethernet/TCP/IP
IDA proposal
Just one word about security
What are the real cost parameters
Conclusion
Friday 28th
September 2001
CERN Workshop Ethernet
5
Vintage Ethernet
Frame broadcasting


Along coaxial cables (10 BAS 5, 10 BAS 2)
Through multiports hubs (10 BAS T, 100 BAS T)
Collision occurrences



Function of the network size, the traffic load, the
average frame size
20% traffic load : 0,1% collisions
40% traffic load : 5% collisions
Non deterministic packet loss and packet delays
Friday 28th
September 2001
CERN Workshop Ethernet
6
Switched Ethernet
10 Mbps, 100 Mbps, 1 Gbps

10 Gbps in the labs !
Frame switching on dedicated ports


On the fly / Store and forward
Half duplex /Full duplex
Friday 28th
September 2001
CERN Workshop Ethernet
7
Level 2
Back Pressure

Operates in half duplex mode
A saturated port sends jam patterns (busy
indicator) to the connected equipment,
enforcing the station to Listen Before Talk

No traffic discrimination

STOP
Friday 28th
September 2001
CERN Workshop Ethernet
8
Level 2
802.3x Flow Control



Operates in full duplex mode
A saturated port sends PAUSE packets to
the connected equipment
No traffic discrimination
PAUSE
Friday 28th
September 2001
CERN Workshop Ethernet
9
Level 2
802.1p CoS



Operates in half or full duplex modes
IEEE 802.1Q defines the general architecture for
Virtual LANs and the VLAN protocol.
IEEE 802.1p defines Classes of Services
Not supported by end-stations adapters
Most switches will not manage the 8 Cos (2 or 4)
Friday 28th
September 2001
CERN Workshop Ethernet
10
Level 3
DiffServ





Differenciated Services
End points managed
Uses IPv4 ToS or IPv6 priority field
64 possible levels
Needs for adequate switches/routers and
adapted applications
Friday 28th
September 2001
CERN Workshop Ethernet
11
TCP/IP
IntServ





Integrated Services
Routers managed
Based on application identification (IPv4
sockets or IPv6 Data Flow Id)
3 levels of services
Uses RSVP Resource reserVation Protocol
to mark a path of reserved ressources for
an application flow
Friday 28th
September 2001
CERN Workshop Ethernet
12
Conclusion on QoS
Management
So many strategies to avoid
or minimize packets loss




IEEE 802.1p : a transient solution ?
DiffServ bind to IPv6 deployment ?
No future for IntServ ?
Products are just coming on the market
Friday 28th
September 2001
CERN Workshop Ethernet
13
Switched Ethernet:
a case study (1/2)
Ethernet Switch
Station




Station
Station
Station
Station
Station
6 stations, cyclically generating realtime 200 bytes
packets, "tagged" high priority
100 Mbps full duplex links
One store & forward switch with an average 10 s
nominal latency
Coexistence with low priority traffic
Compute the worst case total latency for a RT packet
Friday 28th
September 2001
CERN Workshop Ethernet
14
Switched Ethernet:
a case study (2/2)




16 s store & forward
10 s nominal switch latency
122 s flushing (the station can be at this
moment transmitting a low priority max
size frame)
80 s queue emptying (5 RT packets
already in queue)
TOTAL = 228 s
Friday 28th
September 2001
CERN Workshop Ethernet
15
Transmission delays
Yet an another issue !

In complex architectures, end-to-end
delays may be lengthened by
 Gateways
 Proxys
…
An argument for small/average size network
domains
Friday 28th
September 2001
CERN Workshop Ethernet
16
Agenda
Determinism and Qos with Ethernet


"Vintage" Ethernet
Switched Ethernet
 Which QoS implementation ?
An overview of vendors proposals



Fieldbus interconnection with Ethernet/TCP/IP
Fieldbus protocols over Ethernet/TCP/IP
IDA proposal
Just one word about security
What are the real cost parameters
Conclusion
Friday 28th
September 2001
CERN Workshop Ethernet
17
Vendors proposal
first scheme (1/2)
"I offer solutions to interconnect my fieldbus
and Ethernet"
A good idea …



Profinet (Siemens)
WorldFIP
…
Friday 28th
September 2001
CERN Workshop Ethernet
18
Vendors proposal
first scheme (2/2)
But not so simple…




Can I support IP applications on the fieldbus,
without disturbing real time traffic ?
Can I forward fieldbus protocol messages through
Ethernet segments ?
Can I have an integrated solution for the networks
management ?
Delays introduced by proxies and gateway
The vendor must offer a complete solution !
Friday 28th
September 2001
CERN Workshop Ethernet
19
Vendors proposals
second scheme (1/3)
"I take my classic Fieldbus protocol and carry
it on Ethernet with TCP or UDP"
A good idea …





Modbus/TCP (Schneider)
Ethernet/IP (Rockwell, IAONA)
HSE (Fieldbus Foundation)
WorldFIP EtherFIP
…
Friday 28th
September 2001
CERN Workshop Ethernet
20
Vendors proposals
second scheme (2/3)
But not so simple…





TCP/IP protocol stack is designed for Client/server
application, not for 1 producer/N Consumers
exchanges
Needs switched Ethernet to insure determinism
Needs QoS management if mixed traffic with other
IP applications is needed
How manage full redundancy if needed ?
How manage station synchronization if needed ?
The protocol encapsulation specification is not
enough !
Friday 28th
September 2001
CERN Workshop Ethernet
21
Vendors proposals
second scheme (3/3)
For sure not the cheapest solution !


Needs for industrial packaged Ethernet
products
Needs engineering skill if hard real time
constraints
What is the value added for the end
user ?
All this stuff doesn't make the initial fieldbus
protocol better !
Friday 28th
September 2001
CERN Workshop Ethernet
22
IDA proposal (1/2)
Interface for Distributed Automation

NDDS, Network Data Delivery Service
Middleware
 Covering OSI layers 4 to 7
 Producer/Consumer Model
 Broadcast and Multicast
 Adapted to real time traffic
Friday 28th
September 2001
CERN Workshop Ethernet
23
IDA Proposal (2/2)
Programming Interface
UDP
Download
Diagnostics
HTTP
Alarm
SMTP
Event
(T)FTP
Image
Parametrization
SNMP
Process
NDDS
System Management
IDA Object Model
TCP
IP
Ethernet
Friday 28th
September 2001
CERN Workshop Ethernet
24
Conclusion on Proposals




Various solutions…
For various uses ?
Most of them not yet mature
What future for IEEE 1451 ?
Friday 28th
September 2001
CERN Workshop Ethernet
25
Agenda
Determinism and Qos with Ethernet


"Vintage" Ethernet
Switched Ethernet
 Which QoS implementation ?
An overview of vendors proposals



Fieldbus interconnection with Ethernet/TCP/IP
Fieldbus protocols over Ethernet/TCP/IP
IDA proposal
Just one word about security
What are the real cost parameters
Conclusion
Friday 28th
September 2001
CERN Workshop Ethernet
26
Security Issues
Do not mistake Ethernet and Internet !




Consider the plant floor network as an Intranet
Networks manufacturers & designers have the knowhow
There are solutions to secure private IP networks
(I hope that) plant floor networks (will) make use of
them
Do not mistake Security and safety !


Fieldbus manufacturers have the know-how
Mechanisms used to provide safety in classic fieldbus
have to be implemented in Ethernet solutions
Friday 28th
September 2001
CERN Workshop Ethernet
27
Agenda
Determinism and Qos with Ethernet


"Vintage" Ethernet
Switched Ethernet
 Which QoS implementation ?
An overview of vendors proposals



Fieldbus interconnection with Ethernet/TCP/IP
Fieldbus protocols over Ethernet/TCP/IP
IDA proposal
Just one word about security
What are the real cost parameters
Conclusion
Friday 28th
September 2001
CERN Workshop Ethernet
28
Cost parameters (1/3)
Engineering costs


Components standardization
Well-known design rules
Equipement costs


Off-the-shelf components / Customs components
Proprietary protocols / open standards
Installation costs


Number of different networks
Number of cabling systems
Friday 28th
September 2001
CERN Workshop Ethernet
29
Cost parameters (2/3)
Commissioning costs



Operators training
System testing
Proven technology & solid design
Maintenance costs



Various components resellers
Automatic discovery of new components
Avaibility of remote configuration & testing devices
Friday 28th
September 2001
CERN Workshop Ethernet
30
Cost parameters (3/3)
Operating costs


Minimization of operating HMIs
Automatic cold & warm start
Evolution Costs
Friday 28th
September 2001
CERN Workshop Ethernet
31
Agenda
Determinism and Qos with Ethernet


"Vintage" Ethernet
Switched Ethernet
 Which QoS implementation ?
An overview of vendors proposals



Fieldbus interconnection with Ethernet/TCP/IP
Fieldbus protocols over Ethernet/TCP/IP
IDA proposal
Just one word about security
What are the real cost parameters
Conclusion
Friday 28th
September 2001
CERN Workshop Ethernet
32
So, Why Ethernet ?
Alternative vs. The "Fieldbus War" ?


Guarantee of durability
Multiple vendors could decrease costs
Seamless data paths


From the plant-floor to the office
For controllers, PLCs and ERP Systems
One network type


One technical expertise
A global network management
Friday 28th
September 2001
CERN Workshop Ethernet
33
One Network fits all : an
utopia ?
The market today : proprietary solutions




Specific network hardware
Specific protocols
More or less compatible devices
Vendors differences based on technical performances
A challenge for setting a new market ?





COTS network hardware
One standard protocol for RT applications
A large range of Ethernet devices
Vendors differences based on value added services
A new space for systems integrators ?
Friday 28th
September 2001
CERN Workshop Ethernet
34
Questions ?
Friday 28th
September 2001
CERN Workshop Ethernet
35