Re: [PATCH RFC] Support for write stream IDs

[Jens Axboe <axboe@kernel.dk>]

On 03/24/2015 03:46 PM, Ming Lin-SSI wrote:
>> -----Original Message-----
>> From: Jens Axboe [mailto:axboe@kernel.dk]
>> Sent: Tuesday, March 24, 2015 10:08 AM
>> To: Jeff Moyer; Jens Axboe
>> Cc: linux-kernel@vger.kernel.org; linux-fsdevel@vger.kernel.org; Ming Lin-
>> SSI
>> Subject: Re: [PATCH RFC] Support for write stream IDs
>>
>> On 03/24/2015 11:03 AM, Jeff Moyer wrote:
>>> Jens Axboe <axboe@fb.com> writes:
>>>
>>>> Hi,
>>>>
>>>> One of the things that exacerbates write amplification on flash based
>>>> devices is that fact that data with different lifetimes get grouped
>>>> together on media. Currently we have no interface that applications
>>>> can use to separate different types of writes. This patch set adds
>>>> support for that.
>>>>
>>>> The kernel has no knowledge of what stream ID is what. The idea is
>>>> that writes with identical stream IDs have similar life times, not
>>>> that stream ID 'X' has a shorter lifetime than stream ID 'X+1'.
>>>>
>>>> There are basically two interfaces that could be used for this. One
>>>> is fcntl, the other is fadvise. This patchset uses fadvise, with a
>>>> new POSIX_FADV_STREAMID hint. The 'offset' field is used to pass the
>>>> relevant stream ID. Switching to fcntl (with a SET/GET_STREAMID)
>>>> would be trivial.
>>>>
>>>> The patchset wires up the block parts, adds buffered and O_DIRECT
>>>> support, and modifies btrfs/xfs too. It should be trivial to extend
>>>> this to all other file systems, I just used xfs and btrfs for testing.
>>>>
>>>> No block drivers are wired up yet. Patches are against current -git.
>>>
>>> Can you give an idea of how the stream id would be communicated to the
>>> device?  NVMe doesn't appear to have any notion of a data stream ID.
>>
>> It doesn't have it, yet, completely. Ming Lin can expand on what it looks like
>> for the Samsung nvme devices. Current nvme does have vague support for it,
>> however. The write command does have bits for frequent vs infrequent vs
>> one-time writes.
>
> Samsung extended NVMe spec to add "stream control command" to open/close stream.
> And with small modification to the "Write command" to write to open stream.
>
> We are promoting the multi-stream spec to NVMe group.

Can you share the nvme spec proposal with us?

-- 
Jens Axboe