Re: [RFC PATCH] ext3 writepage() journal avoidance

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

Andrew Morton wrote:

>Arjan van de Ven <arjan@infradead.org> wrote:
>
>>On Fri, 2006-03-10 at 00:23 -0800, Andrew Morton wrote:
>> > Arjan van de Ven <arjan@infradead.org> wrote:
>> > >
>> > > 
>> > > > I'm not sure that PageMappedToDisk() gets set in all the right places
>> > > > though - it's mainly for the `nobh' handling and block_prepare_write()
>> > > > would need to be taught to set it.  I guess that'd be a net win, even if
>> > > > only ext3 uses it..
>> > > 
>> > > btw is nobh mature enough yet to become the default, or to just go away
>> > > entirely as option ?
>> > 
>> > I don't know how much usage it's had, sorry.  It's only allowed in
>> > data=writeback mode and not many people seem to use even that.
>>
>> would you be prepared to turn it on by default in -mm for a bit to see
>> how it holds up?
>>
>
>spose so.  One would have to test it a bit first, make sure that it still
>works.  Performance testing with PAGE_SIZE much-greater-than blocksize
>would be needed.
>
I did nobh option only for writeback mode + only if PAGE_SIZE == 
blocksize case :(
I guess I could enhance it for PAGE_SIZE > blocksize case also.

Doing it for ordered mode, journal mode is hard - due to transactions & 
ordering.
As you suggested while ago, we need a new mode. I hate to add new modes 
since
no one will be using it (unless we decide to make it default). Thats the 
reason why
I spent little time doing nobh option for writeback mode.

>
>Unfortunately there's no `-o bh' (nonobh?) to turn it back on again if it
>causes problems..
>

Can be added easily. I will send out a patch for this.

Thanks,
Badari

>