What is it about?
PInTE presents a paradigm of study that's increasingly common and complex to evaluate: running while resources are contended for my multiple applications. The paper presents a method that focuses on a single workload and abstracts contention as random evictions, thus enabling faster, lighter, and tunable cache contention analysis.
Featured Image
Why is it important?
Contention is increasingly prevalent and we cannot simply throw compute at it -- we're running into limits there. We need to study a workload's behavior when under pressure from contention, but running every experiment combination grows the amount of time and computing resources we need. PInTE reduces this cost by abstracting contention as a random event that causes cache pressure.
Perspectives
The methodology was thought of as a solution to the scaling problems we see when running multi-programmed experiments to observe cache contention but can be generalized toward studying the effect of sharing resources (e.g. bandwidth, memory, core cache, etc.).
Dr Cesar Abrao Moody Gomes
Intel Corp
Read the Original
This page is a summary of: PInTE: Probabilistic Induction of Theft Evictions, November 2022, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/iiswc55918.2022.00011.
You can read the full text:
Contributors
The following have contributed to this page
