Investigate considerable memory consumption when collecting code coverage for models
Code coverage for models are extremely useful, especially for debugging purposes. For small jobs it works very well (parallel Weaver helps much), but for large jobs, say, ~2000 drivers of the Linux kernel, Core requires too much memory. For instance, 15 GB is not enough to complete the job. We need to investigate and fix this.
Updated by Evgeny Novikov 11 months ago
- Status changed from New to Rejected
I investigated the issue. Indeed, Core consumes extremely much disk space rather than memory when collecting code coverage for models and when one decides jobs using development settings. Anyway, it is a bad idea to do the latter for any large enough job.
Another observation is that collecting code coverage for models increases the overall walltime in about 50% when checking non CPU intensive requirement specifications. For memory safety it was necessary just ~10% more walltime.