“leaky bucket” style disk io throttling using linux kernel

My goal is to prevent one kvm guest’s disk i/o usage from negatively affecting other kvm guests on the same host. The host has one disk array – and separating guests onto different physical disks is not an option.

qemu 2.6 seems to have this functionality. But I’m running qemu 2.5.

I’m launching my kvm guests using systemd and have IOReadBandwidthMax and IOWriteBandwidthMax set, but since those knobs don’t allow for bursts, it seems like they will simply ensure that all guests have an equally low disk io speeds, rather than preventing one guest from going nuts.

Is there a way to get a “leaky bucket” style algorithm using systemd or directly using cgroups?


Source : Link , Question Author : Jamie , Answer Author : Community

Leave a Comment