UCI P 104 - Methods for Multi Dimensional Robustness Optimization

Unformatted text preview:

Methods for Multi-Dimensional Robustness Optimizationin Complex Embedded SystemsArne Hamann, Razvan Racu, Rolf ErnstInstitute of Computer and Communication Network EngineeringTechnical University of Braunschweig, Germany{hamann|racu|ernst}@ida.ing.tu-bs.deABSTRACTDesign space exploration of embedded systems typically focuseson classical design goals such as cost, timing, buffer sizes, andpower consumption. Robustness criteria, i.e. sensitivity of the sys-tem to variations of properties like execution and transmission de-lays, input data rates, CPU clock rates, etc., has found less attentiondespite its practical relevance.In this paper we introduce multi-dimensional robustness metrics,expressing the static and dynamic design robustness of a given sys-tem, the former assuming a fixed parameter configuration, and thelatter including parameter adaptations as response to property vari-ations. Additionally, we propose a metric measuring the robustnessgain that can be achieved through system reconfigurability.Since determining multi-dimensional robustness is computation-ally expensive we introduce efficient exploration methods based ona stochastic sensitivity analysis technique capable of deriving upperand lower robustness bounds for a given system with low computa-tional effort. We demonstrate the robustness optimization methodsby means of a small but realistic case study.Categories and Subject DescriptorsC.3 [Special-Purpose and application-based systems]: Real-time and embedded systems; C.4 [Performance of systems]: Mod-eling techniques; Performance attributes; Reliability, availability,and serviceabilityGeneral TermsAlgorithms, Design, Performance, Reliability, Verification1. INTRODUCTIONIn the embedded systems design flow, design robustness to prop-erty variations, such as execution and transmission delays, inputdata rates, CPU clock rates, etc., is playing an increasingly impor-tant role.Generally, system robustness is desirable to account for estima-tion errors in early design phases, minor changes of specifications,bug fixes or later extensions and updates of the design to name justa few of the many situations where tolerance of hardware and runtime system to modifications is expected. For instance, it is knownthat small task core execution time modifications in systems withPermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.EMSOFT’07, September 30–October 3, 2007, Salzburg, Austria.Copyright 2007 ACM 978-1-59593-825-1/07/0009 ...$5.00.complex performance dependencies can have drastic non-intuitiveeffects on the overall system behavior, and might lead to severeperformance degradation effects [10].In the current state of practice, designers reserve some slack forcritical system parameters to ensure system robustness. A promi-nent example is the bus load model, where designers limit the av-erage bus utilization to ensure system functionality and extensibil-ity [11].While such design guidelines used to work reasonably well inpractice to ensure design robustness, they are gradually runningout of steam. The main reason for this is the growing size and com-plex networked nature of modern embedded systems, making it ex-tremely difficult to predict the effects of modifications on load andtiming. The complexity of the problem is additionally increasedby the large number of independently developed applications thatare integrated on the same system leading to unknown coupling ef-fects or limitations. Examples are cars or aircrafts. Results are anincreasing design risk and non-extendable systems.In this paper we address the application of formal models to ro-bustness optimization. We first formulate the problem we addressin this paper (Section 2) and give a brief survey of related work(Section 3). Afterwards, we introduce two formal robustness met-rics for different design scenarios (Section 5), which are based onsensitivity analysis (Section 4), and propose efficient explorationmethods considering them during design space exploration (Sec-tion 6). The static design robustness expresses the robustness ofa given system with respect to property variation of a set of criti-cal system properties for a fixed parameter configuration. The dy-namic design robustness additionally includes counter actions (i.e.system parameter adaptations) as response to property variations.Based on the static and dynamic design robustness metrics we in-troduce a metric measuring the robustness gain that can be achievedthrough system reconfigurability. Finally, we demonstrate the pro-posed metrics and techniques by means of a small but realistic casestudy (Section 7).2. PROBLEM STATEMENTThere are many notions of robustness in the context of embed-ded system design. Frequently, robustness is associated with faulttolerance, i.e. techniques that ensure that the system functions cor-rectly in the presence of faults. Examples are task re-execution orreplication mechanisms [8].However, in this paper we define robustness differently. We callsystems robust, if they can sustain property changes (e.g. WCETs,periods, CPU clock rates, etc.) without severe degradation of sys-tem functioning and performance. Consequently, the techniquespresented in this paper are meant to support the designer in con-ceiving systems that are robust with respect to property variations.Basically, we distinguish two different kinds of system propertyvariations: variations influencing the system load, and variationsinfluencing the system service capacity.104Reasons for system load variations are mainly changes of soft-ware execution path lengths, communication volumes, and inputdata rates. Scenarios under which such load variations can occurduring design time or even in the field include late feature requests,product variants, software updates, and bug-fixes.System service capacity variations are caused by modificationsof the execution platform, e.g. processor or communication linkperformance changes. Such variations rarely occur in the field.However, they they are of particular interest during early designspace exploration, where load requirements are still subject tochanges, and where different


View Full Document

UCI P 104 - Methods for Multi Dimensional Robustness Optimization

Documents in this Course
Load more
Download Methods for Multi Dimensional Robustness Optimization
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Methods for Multi Dimensional Robustness Optimization and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Methods for Multi Dimensional Robustness Optimization 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?