Scheduling and Memory requirements analysis with AADL F. Singhoff, J. Legrand, L. Nana, L. Marcé

SIGAda, Atlanta, nov’05 1/20

Scheduling and Memory requirements analysis with

AADL

F. Singhoff, J. Legrand, L. Nana, L. Marcé

University of Brest, France

LISYC/EA 3883

SIGAda, Atlanta, nov'05 2/20

Introduction and motivations (1/2)

The Architecture Analysis & Design Language (AADL) :

• SAE standard published in november 2004 (AS 5506 document).• Allows a high level specification of real time systems in order to perform

analysis, source code/documentation generation, …

An AADL specification is A set of components such as :

• Thread : flow of control that executes a program (eg. Ada Task).

• Data : any data structure in a program (eg. Ada tagged record).

• Processor : part of the execution environment.

Connections between components : model component relationships. Component properties : component information/characteristic

(behavior, Ada package file name, Ada task stack requirement, …)


Introduction and motivations (2/2)

Existing AADL toolsets : OSATE, STOOD, TOPCASED, ADeS

… but there is currently few open-source performance analysis tools.

Since 2002, at Brest, we’re investigating :

• How to extend real time system analysis methods to take distribution into account ?

• How to provide tools to help system designers to predict distributed real time system performances ?

• Cheddar, an Ada framework designed to real time applications performance analysis.

How Cheddar can be applied to AADL specifications ?


Talk overview

Introduction and project motivationsUsual performance analysis methodsCheddar : a performance analyzer frameworkExamples of AADL analysis :

• AADL threads scheduling analysis• AADL event data port memory analysis

Conclusion and ongoing works


Periodic task model : a bound on execution time (capacity, Ci), a fixed delay between two wake-up times (period,Pi), a temporal constraint to meet (deadline, Di).

Choose a real time scheduling algorithms (eg. Rate Monotonic).

Perform simulation analysis : compute and analyze time lines

Perform Analytical/Feasibility test analysis : no scheduling required

Usual performance analysis methods : real time scheduling (1/2)

%69)12( /1

1

nn

i

nPi

Ci


Usual performance analysis methods : queueing systems (2/2)

Queueing system Kendall's notation : X/Y/n.• X : customer arrival rate (M,G,D).• Y : service time rate (M,G,D).• n : number of servers.• Examples : M/M/1, M/D/1, M/G/1, …

Goal : From a given customer arrival/service time rate, compute analytical criteria such as customer waiting time and number of waiting customers in the queue.

Queue ServerCustomer arrivals Outgoing customers


Talk overview





Cheddar : a performance analyzer framework

Cheddar provides Ada 95 packages implementing both:

• Analytical and simulation queueing systems tools (see proceedings of SIGAda 2005).

• Analytical and simulation real time scheduling analysis tools (see proceedings of SIGAda 2004).

Example of computed criteria :

• Processors/tasks : worst/best/average response time, number of context switches/preemptions, missed deadlines, processor utilization factor test, ...• Shared resources & buffers : worst/best/average shared resource blocking task, priority inversion, maximum/average message waiting time, maximum/average number of messages ...• …


Cheddar and AADL

Cheddar was not originally designed to work with AADL. How such kind of tool may be applied to AADL ?

In the sequel, we consider the following examples :

• AADL threads scheduling analysis.

• Buffer/Memory requirements of AADL event data ports.


Talk overview





AADL threads scheduling analysis (1/3)

AADL includes most of the features used in the context of real time scheduling analysis.

Nevertheless, the following questions have to be investigated :

• Can we model any built-in Cheddar's schedulers or task models ?

• Are standard properties enough to perform analytical/feasibility tests on

any resources ?

• How to express new schedulers or task models ?

=> we need some new AADL properties related to thread, data and processor



Example : a set of periodic/aperiodic threads scheduled with POSIX1003.1b and Rate Monotonic schedulers.

thread implementation T3

properties

Dispatch_Protocol => Periodic;

Compute_Execution_Time => 1 ms .. 2 ms;

Deadline => 10 ms;

Period => 10 ms;

end T3;

thread implementation fifo2

properties

Dispatch_Protocol => Background;

Compute_Execution_Time => 1 ms .. 3 ms;

Cheddar_Properties::POSIX_Scheduling_Policy => SCHED_FIFO;

Cheddar_Properties::Fixed_Priority => 5;

Deadline => 100 ms;

end fifo2;

processor implementation cpu0

properties

Scheduling_Protocol => RATE_MONOTONIC;

Cheddar_Properties::Preemptive_Scheduler => true;

Cheddar_Properties::Scheduler_Quantum => 0 ms;

end cpu0;

processor implementation cpu1

properties

Scheduling_Protocol => POSIX_SCHEDULER;

Cheddar_Properties::Preemptive_Scheduler => true;

Cheddar_Properties::Scheduler_Quantum => 2 ms;

end cpu1;



Computesimulation

Analysis(eg. deadlines, response times)


Talk overview

Introduction and project motivationsUsual performance analysis methodsCheddar : a performance analyzer framework Examples of AADL analysis :




Event data ports memory requirement analysis (1/4)

Event data ports are AADL features used for asynchronous message transmission between threads. Events/messages are queued.

Queueing systems may be able to predict event data port me mory requirement .... But we have to answer two questions :

• How to take into account AADL thread dispatching (eg. periodic) ?

• How to take into account AADL thread scheduling (eg. RM) ?

As for the other AADL features, Cheddar provides :

• Simulation tools.

• Feasibility tests (based on queueing systems, J. Legrand, F. Singhoff,

L. Nana & L. Marcé 2004).



Queueing systems for AADL event data ports :• Define a new customer arrival/service time rate : the P rate. • Define new queueing systems based on the P rate.• Define new feasibility tests.

Worst case analytical analysis based on P/P/1 :• Consumer/Producer threads are periodic.• Ex. Feasibility test : the maximum number of messages in an event

data port shared by N periodic producers and 1 periodic consumer is :

• 2.N (harmonic AADL thread set) • 2.N+1 (other cases)

Average case analytical analysis based on M/P/1 :• See the SIGAda 2005 proceedings for the feasibility tests.



processor implementation cpu_rm.i

properties

Scheduling_Protocol => Rate_Monotonic;

...

end cpu_rm.i;

process implementation p0.i

subcomponents

Producer1 : thread Producer.i;

Producer2 : thread Producer.i;

Consumer1 : thread Consumer.i;

connections

event data port Producer1.Data_Source -> Consumer1.Data_Sink;

event data port Producer2.Data_Source -> Consumer1.Data_Sink;

end p0.i;

thread Producer

Features

Data_Source : out event data port;

end Producer;

thread Consumer

features

Data_Sink : in event data port;

end Consumer;

thread implementation Producer.i

properties

Dispatch_Protocol=>Periodic;

...

end Producer.i;

thread implementation Consumer.i

properties

Dispatch_Protocol=>Periodic;

...

end Consumer.i;

Example 2 : event data port connections



Buffer simulation

Analysis fromsimulation

Analyticalworst casequeueing systemanalysis (based on P/P/1)


Talk overview






The Cheddar AADL performance analyzer : • Provides performance analysis tools for different AADL features,

based on real time scheduling and queueing systems analysis.• This AADL analyzer is distributed since the end of october (

http://beru.univ-brest.fr/~singhoff/cheddar) ... but has to be tested !

• Implementation based on Ocarina (ENST’s AADL Ada 95 parser).

• Cheddar can be run as a Stood plug-in, the TNI-Europe CASE tool

(Thanks to AADL !).

Ongoing works :

• Related to thread precedency relationships (AADL connections).

• Related to hierarchical schedulers (eg. ARINC 653 scheduler).

Documents

Scheduling and Memory requirements analysis with AADL F. Singhoff, J. Legrand, L. Nana, L. Marcé