Modelling real-time applications based on resource reservation

Size: px

Start display at page:

Download "Modelling real-time applications based on resource reservation"

Oswin Hawkins
8 years ago
Views:

1 Modelling real-time applications based on resource reservation Laura Barros, César Cuevas, Patricia López Martínez, José M. Drake and Michael González Harbour Grupo de Computadores y Tiempo Real Universidad de Cantabria, Spain {barrosl, cuevasce, lopezpa, drakej, 2 1 nd International Workshop on Analysis Tools and Methodologies for Embedded and Real-time Systems Porto, July 5th, 2011 Santander, 25/09/2010 Funded by the EU under contracts FP7/NoE/214373, and by the Spanish Government under grant TIN C03-03

2 Introduction Resource Reservation (RR) Executing each system thread or communication session in a server Server: it has assigned a fraction of the processor capacity or the communication network. Advantages: System robustness Design simplicity Reusability of software components MAST Open source set of tools to design and analysis of RT applications MAST model extended =>MAST 2 MAST 2 tools under development: Possibility of usage of MAST 1 tools transforming models by MDA 2

Advantages: System robustness Design simplicity Reusability of software components MAST Open source set of tools to

3 Modelling elements for the resource reservation paradigm I e2efa triggera stepa1 ea1 stepa2 ea2 stepa3 tr1 Reactive model e2efb triggerb stepb1 tr2 eb1 stepb2 3

4 Modelling elements for the resource reservation paradigm I e2efa triggera stepa1 ea1 stepa2 ea2 stepa3 Reactive model e2efb triggerb mutexu thread3 stepb1 thread1 tr2 stepb2 processorx primaryscheduler eb1 thread4 thread2 tr1 Resource model 4

e2efb triggerb mutexu thread3 stepb1 thread1 tr2 stepb2

5 Modelling elements for the resource reservation paradigm II e2efa triggera stepa1 ea1 stepa2 ea2 stepa3 tr1 Reactive model virtualrsrc 1 virtualrsrc 2 virtualrsrc 3 virtualrsrc 4 Virtual platform e2efb triggerb stepb1 tr2 eb1 stepb2 5

Reactive model virtualrsrc 1 virtualrsrc 2 virtualrsrc 3

6 New classes in resource reservation MAST models I

7 New classes in resource reservation MAST models II Contract: it represents the capacity required by the application to be executed. VirtualRsrc Attributes of the contract: Budget (t B ) Replenishment period (t R ) Access Deadline (t D ) 7

8 Real-time application development Application Description Timing requirements Application Workload «Mast2» RR-model Application Virtual Platform Design «Mast1» RT_Analysis Model «Mast1» RT_Simulation Model Analysis «Mast2» RR-model & constr. (a) 4d b1 +4d d d b2 +4d d2 +2d a (b) d b1 =3.1, d d1 =8.8, d b2 =9.54,d d2 =7.2,d a2 =1.2 (c)... RR tool MAST tools MAST Simulation tools Negotiate Resource Reservation service Execution platform «Mast2» Exec-model «Mast2» Platform Description Execution Execution platform Application code Asigment VR tool

8 4d b2 +4d d2 +2d a2 101.8 (b) d b1 =3.1, d d1 =8.8, d b2 =9.54,d d2 =7.2,d a2 =1.2 (c).

9 Reactive model of ServoControl example controltrans: EndToEndFlow «PeriodicEvent» period=0.01 virtualplatform T1 clockevent a1 readgoal e1 «Step» SimplOp(wcet=4.0E-4) «Step» message(maxsize=64) Vr_Central (vr1) B = a1 + a5 T = T1 a2 transmitreq e2 «Step» SimplOp(wcet=7.5E-5) a3 readsensorposition e3 a4 returnposition e4 «Step» Message(maxSize=256) «Step» SimplOp(wcet=0.8E-3) Vr_Bus (vr2) B = a2 + a4+a6 T = T1 a5 evaluatecontrol e5 a6 transmitcontrol e6 «Step» Message(maxSize=256) «Step» SimplOp(wcet=5.0E-5) Vr_Remote (vr3) B = a3 + a7 T = T1 a7 setservoinputs end «HardGlobalDeadline» deadline=5.0e-3 refevent= clockevent 9

0E-4) «Step» message(maxsize=64) Vr_Central (vr1) B = a1 + a5 T = T1 a2 transmitreq e2 «Step» SimplOp(wcet=7.

10 Virtual Platform assignment tool Virtual Platform Period Budget Deadline Vr_Central(vr1) T1 a1+ a5? Vr_Bus(vr2) T1 a2+a4+a6? Vr_Remote(vr3) T1 a3+a7? 2 vr1.t D+ 3 vr2.t D+ 2 vr3.t D ms wrt (a x ) < vr.t D (vr.t B a x ) It is possible to calculate the wrt (t X ) of an activity (t a ) in a VirtualRsrc. vr1.t D -(vr1.t B -a1)+ vr1.t D -(vr1.t B -a5)+ + vr2.t D -(vr2.t B -a2)+ vr2.t D -(vr2.t B -a4)+ vr2.t D -(vr2.t B -a6)+ + vr3.t D -(vr3.t B -a3)+vr3.t D -(vr3.t B -a7) < t GD 10

t B a x ) It is possible to calculate the wrt (t X ) of an activity (t a ) in a VirtualRsrc. vr1.t D -(vr1.t B -a1)+ vr1.

11 Negotiation tool Negotiation tool Virtual Platform adjustment Iterative search of a set of deadline values that: makes the application schedulable verifies the specified restrictions Virtual Resources generation MAST Schedulability analysis Tool Vr/SchedRsrc Period Priority Budget Deadline vr1/centrthr T ms 2.51 ms Vr2/commChannel T bits 0.83 ms Vr3/remoteThr T ms 0.5 ms 11

generation MAST Schedulability analysis Tool Vr/SchedRsrc Period Priority Budget Deadline

12 Complex application: No restrictions. Virtual schedulability analysis Model transformation MAST 2 RR-Model Results of analysis using MAST tools vrx:virtualschedulableresource budget=t B replenismentperiod=t R deadline=t D MAST 2 vrxproc:regularprocessor (Default attributes) vrx:thread vrxthload:thread Analyzable-Model Scheduler=vrxSchParam=Fixed PriorityParam(priority=1) Scheduler=vrxSch Param=FixedPriorityParam( priority=2) ATL vrxsch:primaryscheduler Host=vrxProc Policy=FixedPriorityPolicy vrxproc:endtoendflow trg:periodicevent Period=t D act:step Thread=vrxLoad Operation= vrxop vrxop:simpleoperation Wcet= t D -t B

replenismentperiod=t R deadline=t D MAST 2 vrxproc:regularprocessor (Default attributes) vrx:thread vrxthload:thread Analyzable-Model

13 Model Transformation for the MAST 1 compatibility Design MAST 2 MAST 2 RR-Model MAST 2 Schedulability VirtualTool Analysis Result-Model MAST 2 Analyzable-Model ATL MAST 1 MAST-Results MAST 1 Analyzable-Model MAST Schedulability analysis Tool 13

Analysis Result-Model MAST 2 Analyzable-Model ATL MAST 1

14 Conclusions Modelling elements based on RR paradigm with MAST: Scenario: The application can be analysed independently of the current workload The programmers do not require any knowledge about the underlying platform Solution: MAST tools cover the different phases in the development and execution of applications based on resource reservation Relying in the availability of a resource reservation middleware installed in the platform Currently working on: Updating the MAST tools in order to support the new advanced paradigms for real time systems covered by MAST2 Other : Implementation of the virtual platform assignment tool Implementation of the resource reservation service based on Rt-linux 14

Relying in the availability of a resource reservation middleware installed in the platform Currently working on: Updating the MAST tools in order to support the new advanced

Integrating real-time analysis into design flows

Integrating real-time analysis into design flows WATERS 2011 July 2011, Porto, Portugal Michael González Harbour, mgh@unican.es www.ctr.unican.es GRUPO DE COMPUTADORES Y TIEMPO REAL Michael González Harbour