Traffic Measurement for Large Internet Flows Using Multi-stage Landmark LRU Cache

The scalability issue of the Internet management has motivated recent research on partial flow state based traffic measurement and management. Based on our previous Landmark LRU proposal for long-term fast flow detection [Lichang Che, et al. 2005], this paper proposes a novel traffic measurement scheme with partial flow state, Multi-stage Landmark LRU (ML-LRU) cache. Inherited the merit of the Landmark LRU cache in reducing the possibility of large flows being flushed out of the cache, the ML-LRU locks large flows reliably in the cache for traffic measurement. Simulations driven by real Internet traces prove that the ML-LRU scheme uses only one third of the memory space required by the sample-and-hold [Estan, C., et al. 2003] scheme and meanwhile improves measurement performance notably.