Re: [RFC] vhost-blk implementation

[Badari Pulavarty <pbadari@us.ibm.com>]

Avi Kivity wrote:
> On 03/23/2010 04:50 AM, Badari Pulavarty wrote:
>> Anthony Liguori wrote:
>>> On 03/22/2010 08:45 PM, Badari Pulavarty wrote:
>>>> Anthony Liguori wrote:
>>>>> On 03/22/2010 08:00 PM, Badari Pulavarty wrote:
>>>>>> Forgot to CC: KVM list earlier
>>>>>
>>>>>
>>>>> These virtio results are still with a 2.6.18 kernel with no aio, 
>>>>> right?
>>>> Results are on 2.6.33-rc8-net-next kernel. But not using AIO.
>>>
>>> Can you try with aio and cache=none?
>>
>> aio-native,cache=none
>>
>> # time dd if=/dev/vda of=/dev/null bs=128k iflag=direct
>> 640000+0 records in
>> 640000+0 records out
>> 83886080000 bytes (84 GB) copied, 470.588 seconds, 170 MB/s
>>
>
> You're effectively killing readahead with this.  

Yes. cache=none, aio=native will force it hit the disk all the time. We 
loose all the benefits of
readahead.
> Please use fio with a sequential pattern, queue depth > 1.
>
Will do.

Thanks,
Badari