D. Schmidt DARPA DARPA Current Adaptation responses D istributor I P B I I I I I I I I P P P P P P P P P P B B B B B B B B B B B B B B B B BB B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B B BBBBBBBBBBBBBBBBBBBBBBBBBBBB PPPPPPPPPP IBBPBBPBBPBBPBBPBBIBBPB NETWORK RESERVATION •Under excessive Network load - Use IntServ to reserve bandwidth D istributor ...PBBPBBPBBI I I I DATA FILTERING •Excessive network or CPU load - Drop selective frames LOAD BALANCING •Excessive CPU load - Move distributor to more lightly loaded host X X Variations in Operating Conditions Variations in Mission Requirement s Timeliness •Pilot or targeting officer must have an out- of-the-window view of UAV imagery Quantity •Surveillance officer must record complete UAV imagery for off-line analysis
Transcript
D. Schmidt
DARPADARPACurrent Adaptation responses
Distributor
I PB
II
IIIIIIP
PP
PPP P P
P
P
B
B BBBBB
B BBBBBBB
BBBBBBBBBBBBBBBBBBBBBBB
BBBBBBBBBBBBBBB
BBB BBBBBBBBBBBBBBBBBBBBBBBBBBBBBP P P P P P P P P PIBBPBBPBBPBBPBBPBBIBBPB
NETWORK RESERVATION•Under excessive Network load - Use IntServ to reserve bandwidth
Distributor...PBBPBBPBBI I I IDATA FILTERING•Excessive network or CPU load - Drop selective frames
LOAD BALANCING•Excessive CPU load - Move distributor to more lightly loaded host
X X
Variations inOperating Conditions
Variations inMissionRequirements
Timeliness•Pilot or targeting officer must have an out-of-the-window view of UAV imagery
Quantity•Surveillance officer must record complete UAV imagery for off-line analysis
Adaptive feedback loops can run at multiple layers
Distribution Middleware QoS
Network QoS
Common MiddlewareService QoS
System-wide QoS
Operating System QoS
Application or Domain-specific QoS UAV capabilities CEC/SIAP
Hybrid & MultiChannel
D. Schmidt
DARPADARPA
Contract
QoS Adaptive Control
Contract
Display RateCPU load
Current Layers of QoS Specification& Adaptation in Navy UAV Systems
Network QoS
Common MiddlewareService QoS
Operating SystemQoS
Application or Domain-specific QoS
DistributionMiddleware QoS
VideoForwardingCode
VideoSource
VideoDisplayHyper-D
ResourceManager
Contract
Display RateCPU load
ACE/TAO RT ORB
ACE/TAO RT ORB ACE/TAO RT ORB
IntServ/RSVP
AQoSA
IntServ/RSVP
AQoSA
QoS Adaptive Control
QoS Adaptive Control
A/V Streaming Service A/V Streaming Service
D. Schmidt
DARPADARPAAdHoc Integration of Components for
QoS Adaptation and Control
VideoForwardingCode
VideoSource Video
Display
DistributedResourceManagementCoordination
ACE/TAO RT ORB
A/V StreamingService
DataPath
ACE/TAO RT ORB
LANNetwork
OperatingSystemOperating
System
WirelessNetwork
ControlPath
Contract
QoS Adaptive Control
DataPath
ControlPath
ACE/TAO RT ORB
A/V StreamingService
Contract
QoS Adaptive Control
Contract
QoS Adaptive Control
A/V StreamingService
AQoSAAQoSA
IntServ/RSVP IntServ/RSVP
AQoSA
VirtualInformationCollection
IntServ/RSVP
OperatingSystem
D. Schmidt
DARPADARPAUAV/HIPER-D Requirements
(Previous Experiment)
Low latency to support interaction (users see images at the same time as the UAV)
Displayed frame rate can be less than 30/second, providing that targets remain clear and no jitter
HIPER-D Resource Manager determines where and when applications run
Management techniques focused on discrete problem and remedy
New experiments will extend these basic ideas:– Individual and composite bottleneck identification and adaptations– End-to-end behavior– Aggregate and Coordinated behavior– Scaling and Redundancy– Varied anomalies and operating conditions– More resources under control/coordination, including “soft” resources– Intercluster coordination and feedback
D. Schmidt
DARPADARPA
UAV demonstration illustrates some of the software engineering challenges with reusing and adding QoS
to current off-the-shelf component software We used an off-the-shelf video player in the UAV demonstration
– Developed for playing MPEG video from a file
– Had to convert it to accept input from a stream
Developers of the video player had recognized the need for adaptation to handle changes in QoS
– The video player included code to compensate for slow video cards (i.e., falling behind in the video)
– Unfortunately, this code is intertwined throughout the functional code (i.e., there is no separation of concerns)
Reusing this code presented some challenges because the QoS code was intertwined and specific to a different use-case
– We had to “turn off” the file-specific adaptation in order to use the video player effectively with a video stream
– This was difficult because the adaptive code was intertwined throughout the functional code
In contrast, the adaptive code specified separately in the QuO middleware was easy to change
D. Schmidt
DARPADARPA
Client Object
ORB endsystem
ORB endsystem
ResourceResource
Resource
Applying Reflection to Optimize Multi-level Resource Management
Key System Characteristics
• Integrate observing & predicting of current status & delivered QoS to inform the meta-layer
•Meta-layer applies reflection to adapt system policies & mechanisms to enhance delivered QoS
QoSDoctrine
Applying reflection as an optimization is even more relevant to middleware than compilers due to dynamism & global resources:
PiggybackedMeasurements
ExpectedQoS
MeasuredQoS
CorrelateProbes
Status
Resource Management Service
CollectTranslateIntegrate
Infer/AdaptFeedback
Loop
Interceptor Interceptor
Interceptor Interceptor
Monitors Monitors
Monitors Monitors
D. Schmidt
DARPADARPAKey Research Challenge:Providing & Organizing QoS Guarantees for Multiple Adaptive Feedback Loops
