Steve Thompson wrote: > On Wed, 11 Apr 2012, m.roth at 5-cent.us wrote: > >> Been seeing that on 6.2, also, though I've not noticed them locking up >> (the ones that have were in a cluster, and someone could have hosed >> memory). It got less with the latest kernel, but I'm still seeing them >> occasionally. > > This is most likely a problem due to data arriving too quickly for the VM > subsystem to flush dirty pages to disk to keep up; the kernel switches to > synchronous mode when the dirty_ratio percentage of memory pages is used. > You might try upping your dirty_ratio, or reducing it drastically. > But mine are *not* VMs. mark