Integration with AbsInt TimeWeaver

Hybrid WCET analysis with TimeWeaver, the timing analyzer from AbsInt: they supply the static analysis, we supply measured times from real silicon.

TimeWeaver is the hybrid WCET analyzer of AbsInt. Our measurements reach it as follows.

Multi-core WCET evidence usually combines architectural analysis, controlled scenarios and measurements, because interference depends on shared resources and concurrent activity. The required balance is method- and platform-specific.

Accemic observes execution on the target and supplies identified path and timing observations. TimeWeaver performs its stated timing-analysis functions on the supplied program and measurement information. The interface does not convert an observed maximum automatically into a proven upper bound; the resulting claim is determined by the TimeWeaver analysis, its assumptions and the project evidence.

Why hybrid

Pure measurement finds the worst case you happened to hit. Static analysis derives an upper bound from an abstract model; where cache, pipeline or shared-resource behavior cannot be modeled tightly, that bound can become conservative - in some systems far enough above observed behavior that it stops being useful for dimensioning.

Hybrid analysis takes the structure and the guarantees from static value and path analysis and the per-block execution times from measurement on the real part. AbsInt TimeWeaver does the analysis; CEDARtools^® supplies the measurements.

What the measurement side does differently

Conventional trace tooling records the raw stream into a capture buffer and decodes it afterwards. At trace rates of several GiB/s even for medium-sized processors, the buffer fills quickly - and the observation ends there. A worst case that needs several subsystems to align may simply never fall inside that window.

CEDARtools inverts the order. The trace stream is decoded live, inside an FPGA, at the speed the target produces it - nothing is stored first, so the length of the observation is not bounded by a capture buffer. A rule engine watches the reconstructed control flow while the test executes: each time a task exceeds its previously observed maximum execution time, a trigger fires and the raw trace of that run replaces the earlier one in a re-triggerable ring buffer, one slice per task, for up to 32 tasks and interrupt handlers in parallel.

Continuous observation with live triggering: the trace stream is decoded as it arrives, a trigger fires when a task exceeds its previous maximum, and a re-triggerable ring buffer keeps the raw trace of the worst observed run of each task - up to 32 in parallel - as input for TimeWeaver's hybrid WCET analysis.

The observation can therefore run for hours, or days: its duration is not bounded by the capacity of a raw-trace capture buffer, subject to continuous trace delivery and validity monitoring. The longer it runs, the more opportunity rare scenarios have to occur - observation length raises exposure, not by itself representativeness. At the end of the campaign the buffer holds the evidence worth keeping: the worst observed run of each task, as raw trace, together with its campaign context, ready for import into TimeWeaver as measurement input to the hybrid analysis. All of this happens without instrumenting the application: the stream comes from the processor’s embedded trace unit, not from added measurement code, so the code under measurement runs unchanged.

The comparison from the joint ERTS² 2024 paper: with a finite capture buffer (top), the observed span between best-case (BCET) and worst-case (WCET) execution time covers only what happened to fall into the buffer window; with continuous observation and pre-qualified snippets (bottom), the rare extremes have time to appear. Blue squares are recorded snippets, gray ones were observed and skipped.

The pipeline and the comparison above come from the joint ERTS² 2024 paper with AbsInt and the University of Lübeck, Multi-core WCET Analysis Using Non-Intrusive Continuous Observation. The paper is available on our publications page or directly in the HAL archive.

What Accemic contributes

Timing observed over long-duration, continuous observation - hours, or days during a test campaign - rather than a short capture window, subject to continuous trace delivery and validity monitoring. Intermittent worst cases that only appear under a specific interference pattern have a chance of being seen, because the observation does not stop.

The transfer record identifies the executable, core and task context, observation interval, clock basis, trace gaps, timing correlation, scenario or workload, and the TimeWeaver project to which the data were imported.

Certification context

AbsInt maintains qualification support kits for DO-178C / DO-330 up to DAL A, ISO 26262 up to ASIL D, IEC 61508 SIL 4 and EN 50128/50657. That is their decades of investment, not ours, and it is the reason the integration runs in this direction.

The combined workflow can contribute measured execution and interference observations to a project argument addressing the objectives of EASA AMC 20-193. Sufficiency depends on the system architecture, interference analysis, test strategy, coverage of relevant operating conditions and acceptance by the responsible authority.

Buying it

TimeWeaver and TimingProfiler are available bundled with CEDARtools. Ask for details.

TimeWeaver, TimingProfiler, Astrée and CompCert are products of AbsInt Angewandte Informatik GmbH. AbsInt and Accemic are not corporately affiliated; neither the integration nor the bundled offering transfers qualification or assurance status between the products.