Download - Adaptive Resource Management Architecture for DRE Systems Nishanth Shankaran [email protected]
Adaptive Resource Management Architecture for DRE Systems
Nishanth ShankaranNishanth [email protected]@dre.vanderbilt.edu
2
Motivation: Distributed Real-time & Embedded Systems
ProblemProblem• Need to operate in open & unpredictable
environments• No accurate apriori knowledge of operating
conditions, resource availability, and input workload• Effective utilization of multiple resources –
computational power and network bandwidthSolutionSolution Adaptive Resource Management
Architecture – Resource Allocation and Control Engine (RACE)
System CharacteristicsSystem Characteristics
• Operate under limited resources
• Tight real-time performance QoS constraints
• Dynamic & uncertain environments
• Loss or degradation of hardware with time
• Distribution of computation
• Multiple nodes & data centers
• Task distribution among hosts/data centers
• Integration of information
• Data collection – Radar
• Compute counter measure(s)
• Execute counter measure(s)
• Coordinated operation• E.g., NASA Earth Science Mission &
Total Ship Computing Environment
3
Resource Allocation and Control Engine
• Dynamic resource management framework atop CORBA Component Model (CCM) middleware (CIAO/DAnCE)
• Allocates components to available resources
• Configure components to satisfy QoS requirements based on dynamic mission goals
• Perform run-time adaptation
• Coarse-grained mechanisms
• React to new missions, drastic changes in mission goals, or unexpected circumstances such as loss of resources
• e.g., component re-allocation or migration
• Fine-grained mechanisms
System Designer/Mission Planner
Target Platform with Varying Resource Availabilities and Capabilities
Resource Allocation & Control Engine (RACE)
Allocation Algorithms
Control Algorithms
Operational Strings with Varying Resource & QoS Requirements
Uniform Interface to Deploy and Manager
Components
CIAO/DAnCE Middleware
Deploy on Target Domain
Monitor and Adapt Resource Allocation
to Application Components
• Compensate for drift & smaller variations in resource usage
• e.g., adjustment of application parameters, such as QoS settings
4
DRE System Model
Client QoS Settings
Server QoS Settings
QoS Enabled Network
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Realtime Distribution Middleware
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Client Proxy
Server Proxy
Client Server Application Parameters
Middleware QoS
Parameters
OS Priorities, Processor
Reservation Quota
Bandwidth Reservation Mechanism /
Diffserv Codepoints
Qu
alit
y o
f S
ervi
ce S
etti
ng
s
QoS Setting at the Application
Layer
QoS Setting at the Middleware
Layer
QoS Setting at the OS Layer
QoS Setting at the N/W Layer
QoS parameters are all layers need to be configured/managed to met end-to-end QoS requirements
5
System Model of a CCM Based DRE System
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
RT QoS Enabled CCM Middleware (CIAO/DAnCE)
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Stub SkeletonRT CCM
Parameters
Linux RT Kernel /
Resource Kernel
Bandwidth Broker
Container
Executor
Container
Executor
RT CCM Policy
Application Parameters
Maps To
RACE
QoS Enabled Network
Qu
ali
ty
of
Se
rvic
e S
ett
ing
s
Container provides an
encapsulation for the application
QoS settings are specified at the container level
These settings are then used to
configure the middleware
RACE currently manages OS QoS parameters/knobs
to meet e-2-e QoS requirements
Bandwidth Broker determines Network
QoS settings
6
System Model of a DDS Based DRE System
DDS QoS Parameters
Linux RT Kernel /
Resource Kernel
Bandwidth Broker
Application Parameters
RACE
Qu
ali
ty
of
Se
rvic
e S
ett
ing
s
Publisher QoS
Settings
Consumer QoS
Settings
QoS Enabled Network
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Data Distribution Service
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Publisher Proxy
Consumer Proxy
Publisher Consumer
RACE can manage OS & N/W QoS settings even for DDS based systems
7
Concluding Remarks and Future Work
• Architecturally, both distribution middleware are similar
• Resource/QoS management architecture developed for one can be applied for the other with minor modifications
• Currently, we have applied RACE for CCM based DRE systems
• In the future, we plan to apply RACE for DDS bases DRE systems
Publisher QoS
Settings
Consumer QoS
Settings
QoS Enabled Network
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Data Distribution Service
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Publisher Proxy
Consumer Proxy
Publisher Consumer
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
RT QoS Enabled CCM Middleware (CIAO/DAnCE)
OS Kernel
RT SchedulerRT CPU Reservation
Mechanism
RT Network Subsystem
Stub Skeleton
Container
Executor
Container
Executor
QoS Enabled Network