adaptive and reconfigurable communication for distributed embedded systems adaptive and...
Post on 18-Dec-2015
232 views
TRANSCRIPT
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Adaptive and reconfigurable Adaptive and reconfigurable communication for distributed communication for distributed
embedded systemsembedded systems
Luís [email protected]
Embedded Systems Lab - IEETA / DETIUniversidade de Aveiro
Aveiro, Portugal
ArtistDesign workshop on adaptivity, 2008-May-13,14ArtistDesign workshop on adaptivity, 2008-May-13,14
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 22
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
What we want
To be able to connect any component to the system, on-line, being sure that:
Nothing bad will happenThe system will do its best to integrate the
added component: It can accept the new component without any
adjustment on the system It can accept it upon system adjustment It can reject the new component
Recon
figur
abilit
y
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 33
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
What we want
Allow the system to adjust on-line according to effective instantaneous needs or resource availability:
Free and reuse the resources of subsystems that operate occasionally / fail when off
Adapt the resources required by each subsystem on-line to:
Minimize the resources used (e.g. to minimize BW usage, energy, …)
Maximize the service delivered with a fixed level of resource usage
Cope with varying levels of resource availability
Adapt
ivity
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 44
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Two approaches
Law enforcementNegotiate needs and enforce their proper useRequires good level of control over resources
Social groupsWork within framework imposed from outsideDo the best possible with what is available
Best effort
Hard real-time
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 55
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Outline
• FTT framework – new developments• FTT-SE (enhanced aperiodic communication)• Dynamic QoS management (distributed video system)• Server-SE (server-based communication management)• Enhanced switch (boosting robustness and integration capab.)
• Wireless communication for teams of robots• Adaptive-TDMA framework• Round reconfiguration
Flexible Time-Triggered architecture
Law enforcement
Social groups
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 66
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
FTT-SEFlexible Time-Triggered communication over Switched Ethernet
Keeping under control the traffic load submitted to a switched network
Schedule traffic per cycles Submit only the traffic that fit in a cycle Eliminate memory overloads Support full priority scheduling FTT master
TM
Trigger message
Ethernet switch
TM
SRTA(...)B(...)
...
sched
FP, EDF, ...
M nodes
Switch with M portsFTT master
lujldj
SMi
Broadcast to all nodes
Nodes reaction to the TM in a given EC
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 99
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
FTT-SE• Same triggering for all traffic
– Aperiodic traffic is signalled to the Master– All traffic scheduled in an integrated way
• Synchronous + asynchronous RT + Non-RT• Everything encoded in the TM
TM
Signaling message
Sched.
AM arrivesAM polled
AM transmitted
Lsig Lpoll
Aperiodic signalling mechanism
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1010
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
QoS adaptation with FTT-SE
FTT master
TM
Trigger message
CBR channels
MJPEG encoders – VBR bit streams
Two dimensional problem• Adaptation of a VBR source to a CBR channel• Adaptation of the bandwidth of the CBR channels
– exploit BW released by streams that are off– To reduce the use of too strong compression
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1212
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
A distributed monitoring system
• Adapting multiple CBR channels– Streams are not always ON– Maximize total BW usage
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1313
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
A distributed monitoring system
• Adapting multiple CBR channels– Evolution of the Quality Index (QI)
comparing to statically alocated channels
2222
2
ˆˆˆˆ2
gf
gf
gf
fg
gf
gfQI
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1414
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Server-SE• All aperiodic
– Uses aperiodic mechanism of FTT-SE– All traffic handled through servers
• Servers controls encoded in the TM– Aims at dynamic managing of the servers
Stream with randomly arriving requests
Sporadic server with one packet every 5 ECs
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1515
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Resource-reservation switch
Providing timeliness, flexibility and high robustness in switched Ethernet networks
Enforce negotiated channel characteristics (policing) Reject abusive negotiated traffic (filtering) Confine non-negotiated traffic to separate windows (selection)
RR-switchRR-switch
RR-switch
Internet browser
Internet
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1616
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Resource-reservation switch
• Aperiodic traffic confinement
Submitted traffic1000B packets, Tavg=250s
Outgoing trafficOffset wrt the previous TMAsync window - 50% EC
• Regularity of the TMT_TMavg = 1,000msT_TMmax = 1,0003msT_TMmin = 0,99998msSTD_TM = 138ns
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1717
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Robust IEEE802.11 communication
Providing robust IEEE 802.11 communication for teams of autonomous mobile robots
Real-time is best-effort Open medium, uncontrolled environment / load,
non-stationary interference... Adaptive techniques can help reducing chances of
packet losses
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1818
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Robust IEEE802.11 communication
Adaptive TDMA Maximizes separation between transmissions in the team Synchronizes on receptions (no need for clock sync) Shifts phase of TDMA round to match periodic interference Time constraints round period Ttup
Fully distributed
tnow
Txwin
M1,i M2,i M3,i M4,i
M1,i+
1
δ
4
Txwin
Ttup
M1,i M2,i M3,i M4,i
M1,i+
1
δ
4
tnext
Non-synchronized periodic broadcast
Adaptive TDMA
(less losses)
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 1919
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Robust IEEE802.11 communication
Dynamic reconfiguration of the slot structure Robots join and leave dynamically
crash, maintenance, movements... Slot structure of TDMA round does not need to be
predefined Number of slots continuously adjusted to number of robots
Fully distributed – minimal a priori knowledgeonly 1 robot
2nd robot joins
3rd joins
2nd leaves
May 13-14, 2008May 13-14, 2008 ArtistDesign workshop for adaptivity, Lund, SwedenArtistDesign workshop for adaptivity, Lund, Sweden 2020
Adaptive and Reconfigurable Communication for Distributed Embedded SystemsAdaptive and Reconfigurable Communication for Distributed Embedded Systems
Conclusion
Adapting / reconfiguring a distributed system on-line requires appropriate support from the network
Two main approaches can be followed:
Law enforcement Such as the FTT framework, aiming at controlled
adaptation/reconfig. (timeliness guarantees, traffic isolation…)
Social groups Wireless communication for teams of autonomous agents,
using of adaptive techniques can help reduce packet losses in the presence of uncontrolled load (best effort way)
Which to use?! How much control have we got over system resources?