Hi all,
I have a KVM Debian lenny machine running on an up to date proxmox 1.5, 2.6.24 kernel with a RAW file mapped as an IDE file (/dev/hda).
It is a postgres server and is *really* slow until you turn on DMA (hdparm -d1 /dev/hda). The problem is that after a while it goes slow again and it turns out that DMA has been turned off. dmesg contains a few entries of:
Has anyone experienced this before? What does this mean - is dma unreliable in KVM? If so, what can I do about it as the performance is terrible without it? As a non-scientific benchmark, rendering the JIRA home page without DMA takes 10 seconds or so, it is almost instant with DMA enabled.
Would reverting back to a 2.6.18 kernel resolve this issue? I cannot move to the .32 kernel as I need the containers.
I have a KVM Debian lenny machine running on an up to date proxmox 1.5, 2.6.24 kernel with a RAW file mapped as an IDE file (/dev/hda).
It is a postgres server and is *really* slow until you turn on DMA (hdparm -d1 /dev/hda). The problem is that after a while it goes slow again and it turns out that DMA has been turned off. dmesg contains a few entries of:
Code:
[36400.000040] hda: dma_timer_expiry: dma status == 0x21
[36410.000041] hda: DMA timeout error
[36410.004022] hda: dma timeout error: status=0xd8 { Busy }
[36410.004022] ide: failed opcode was: unknown
[36410.004022] hda: DMA disabled
[36428.256238] ide0: reset: success
Has anyone experienced this before? What does this mean - is dma unreliable in KVM? If so, what can I do about it as the performance is terrible without it? As a non-scientific benchmark, rendering the JIRA home page without DMA takes 10 seconds or so, it is almost instant with DMA enabled.
Would reverting back to a 2.6.18 kernel resolve this issue? I cannot move to the .32 kernel as I need the containers.