On Fri, Dec 07, 2012 at 10:43:24AM +0800, Liu Yuan wrote: > On 12/06/2012 04:46 AM, Stefan Hajnoczi wrote: > > Limitations: > > * Only format=raw is supported > > The boost number looks promising, but support of other format(which > might as well fit into this new IO path design) is in the plan? It seems > that bypassing block layer would end up adding yet another 'block layer' > as we add more format support.
Right, that's why the next step is to complete the AioContext work that Paolo recently contributed. It will allow us to run the QEMU block layer outside the global mutex. Once that refactoring is complete the data plane thread will be able to run image format code. We'll need to be careful so as not to lose the level of performance that it achieves right now, but there is no fundamental reason why we cannot continue to keep this level of performance while still supporting image formats. Stefan
