Researchers have introduced “LAVA: Lifetime-Aware VM Allocation with Learned Distributions and Adaptation to Mispredictions”, a system designed to improve how cloud data centers place virtual machines or VMs on physical servers. The work tackles the bin packing problem that occurs when VMs appear and disappear with unknown lifespans, making it harder to use server capacity efficiently.
The source text says inefficient VM allocation can cause “resource stranding”, where leftover server resources are too small or unbalanced to host new VMs. It can also reduce the number of “empty hosts”, which are needed for system updates and for provisioning large, resource-intensive VMs.
LAVA includes a trio of algorithms: non-invasive lifetime aware scoring (NILAS), lifetime-aware VM allocation (LAVA), and lifetime-aware rescheduling (LARS). The system is based on “continuous reprediction”, which means it does not rely only on a single lifespan estimate made when a VM is created.
Instead, the model updates its prediction for a VM’s expected remaining lifetime as the VM continues to run. The source text says this is meant to address the problem of mispredictions, where a single wrong forecast can tie up an entire host for an extended period and reduce efficiency.
The announcement frames the work as a response to the challenge of allocating processing jobs in large data centers, where efficient resource use matters for both economic and environmental reasons.
Source: research.google.
Companies can share verified announcements through Newz9’s international press release submission page.
