Most previous work on virtual machine (VM) consolidation assumed that VMs residing on the same host were completely isolated from each other. But this is wrong, as Kim et al. convincingly demonstrate: the shared last-level cache and memory interface may lead to contention between co-located VMs, thus potentially resulting in performance degradation.

The authors devise a simple model for predicting the performance interference that can be expected when colocating a pair of VMs. As shown by empirical results, this model can estimate performance degradation of real-world applications fairly well. Moreover, the authors present a novel heuristic algorithm for VM placement based on the estimated performance interference among the VMs.

The main strength of the paper is that it bridges the gap between problems that are usually handled by different research communities: cache management, virtualization, and VM placement. The paper clearly demonstrates that an integrated approach is indeed necessary for achieving good results, and it makes a good first step toward such an approach. It provides a basis for future research to further improve estimation accuracy as well as VM placement decisions based on the performance interference estimations. The assumptions underlying the work, for example, each VM is mapped to a single core of the host, are well documented; it is also an interesting direction for future research to determine how those assumptions can be relaxed.