Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Just out of the blue: could the problem be in the drbd vs. md interaction, rather than SATA, DMA, IRQ magic? Did anyone ever successfully run drbd on top of md (and under significant load: the problem does not show up under light load)? I remember Philip told me that linbit is typically running drbd on top of hardware RAIDs, and that they are *considering* software RAID because of better monitoring features. And I remember Lars recently said that they failed with software RAID on SATA but things magically went good when they dropped in "an old promise controller" (which is a hardware RAID, correct?). Could the FS corruption be an artifact of "double virtualization" conflicting with the scheme of marking dirty buffers on the VM layer or something equally incomprehensible? ;-) Eugene