1270-GRP Can the Filter Be Gamed? Attacking and Hardening S3-FIFO and S4-FIFO Cache Eviction

Presenter Information

Location

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Document Type

Event

Start Date

20-11-2026 2:00 PM

Description

S3-FIFO and its 2026 successor S4-FIFO admit objects to the main cache through a small FIFO filter or a ghost queue. We show that a client sharing the cache can game both paths. On five public traces and a synthetic workload, burst-repeated requests bypass S3-FIFO’s filter, ghost replay hurts both algorithms, and attack traffic steers S4-FIFO’s learned controller. We propose a ghost-promotion budget, a 62-line change to libCacheSim. Combined with S4-FIFO’s skip ratio, it cuts mean burst-repeat damage from 3.86 to 0.26 percentage points and ghost-replay damage from 5.21 to 2.93, at a mean benign cost of 0.02.

This document is currently not available here.

Share

COinS
 
Nov 20th, 2:00 PM

1270-GRP Can the Filter Be Gamed? Attacking and Hardening S3-FIFO and S4-FIFO Cache Eviction

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

S3-FIFO and its 2026 successor S4-FIFO admit objects to the main cache through a small FIFO filter or a ghost queue. We show that a client sharing the cache can game both paths. On five public traces and a synthetic workload, burst-repeated requests bypass S3-FIFO’s filter, ghost replay hurts both algorithms, and attack traffic steers S4-FIFO’s learned controller. We propose a ghost-promotion budget, a 62-line change to libCacheSim. Combined with S4-FIFO’s skip ratio, it cuts mean burst-repeat damage from 3.86 to 0.26 percentage points and ghost-replay damage from 5.21 to 2.93, at a mean benign cost of 0.02.