Re: [PATCH, 3.7-rc7, RESEND] fs: revert commit bbdd6808 to fallocateUAPI

[Eric Sandeen]

On 12/7/12 3:57 PM, Chris Mason wrote:
> On Fri, Dec 07, 2012 at 02:49:04PM -0700, Ric Wheeler wrote:
>> On 12/07/2012 04:43 PM, Chris Mason wrote:
>>> On Fri, Dec 07, 2012 at 02:27:43PM -0700, Theodore Ts'o wrote:
>>>> On Fri, Dec 07, 2012 at 04:09:32PM -0500, Chris Mason wrote:
>>>>> Persistent trim is what I had in mind, but there are other ideas that do
>>>>> imply a change in behavior as well.  Can we safely assume this feature
>>>>> won't matter on spinning media?  New features like persistent
>>>>> trim do make it much easier to solve securely, and using a bit for it
>>>>> means we can toss back an error to the app if the underlying storage
>>>>> isn't safe.
>>>> We originally implemented no hide stale for spinning media.  Some
>>>> folks have claimed that for XFS their superior technology means that
>>>> no hide stale doesn't buy them anything for HDD's.  I'm not entirely
>>>> sure I buy this, since if you need to update metadata, it means at
>>>> least one extra seek for each random write into 4k preallocated space,
>>>> and 7200 RPM disks only have about 200 seeks per second.
>>> True, 7200 RPM disks are slow, but even allowing them to expose stale
>>> data just makes them a little less slow.
>>>
>>> I know it's against the rules to pretend that disks don't matter.  But
>>> really, once you're doing random IO into a spindle you've given up on
>>> performance anyway.
>>>
>>> -chris
>>
>> That's right.
>>
>> And equally true, once you have moved the disk heads to that track, you can 
>> write a lot as cheaply as a little (i.e., do 1MB instead of 4KB). That will also 
>> avoid fragmentation of the extents.
> 
> When you do a 4K write, you have to remember that you've written just
> those 4K.  When you do a 1MB write, you have to remember that you've
> written just that 1MB.  It's the same operation, except with the 1MB
> you've also had to setup all the bios and send down the zeros, and do
> the proper locking to make sure you're not sending zeros down over
> some concurrent IO.
> 
> The 1MB setup is actually more work, but it does greatly reduce the
> amount of time the workload needs to run before it goes into a steady
> state.  For smaller files it may work well, but for larger ones I don't
> think it will be enough.

Ext4 already does this, actually, I think - see s_extent_max_zeroout_kb
and how it's used.

        /* If extent is less than s_max_zeroout_kb, zeroout directly */

It's not a tunable (*gasp* ;)) but it's currently set to "32" as in
32 kb.  Would be fun to bump that up and see how your test goes.

-Eric

> -chris
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/