2006-02-13 09:54:37

by Victor Porton,,,

[permalink] [raw]
Subject: MTD for buffered IO

I suggest to add kernel configuration/option to allow to use
an MTD device as a continuation of I/O buffers (for HDD).

One way to implement it would be to add the option for a swap partition/file to
allow to use this swap partition/file as I/O buffer for other device.


2006-02-13 13:39:43

by Josh Boyer

[permalink] [raw]
Subject: Re: MTD for buffered IO

On 2/13/06, Victor Porton,,, <[email protected]> wrote:
> I suggest to add kernel configuration/option to allow to use
> an MTD device as a continuation of I/O buffers (for HDD).
>
> One way to implement it would be to add the option for a swap partition/file to
> allow to use this swap partition/file as I/O buffer for other device.

I'm not quite sure what you mean here, but suggestions in the form of
patches go a lot further than just an email.

josh

Subject: Re: MTD for buffered IO



> Victor Porton,,, <[email protected]> wrote:
>
> > I suggest to add kernel configuration/option to allow to use
> > an MTD device as a continuation of I/O buffers (for HDD).
>
> Why do you want to be limited to MTD devices?
> What about creating a device-mapper target using JBD instead?

Good idea, which probably already works with current Linux with ext3. But we should be not limited to JDB systems. Your idea works only with JDB, so your idea (despite of being a good idea) is a hack.

> > One way to implement it would be to add the option for a swap
> partition/file
> > to allow to use this swap partition/file as I/O buffer for other device.
>
> -EBADIDEA. You'd certainly lose data on power failures.

How it is worse than to lose data on power failures with the standard Linux memory I/O cache?

Moreover so we can lose LESS, as Flash memory will preserve swap content unlike RAM.

> --
> Ich danke GMX daf?r, die Verwendung meiner Adressen mittels per SPF
> verbreiteten L?gen zu sabotieren.
>

--
Victor Porton ([email protected])

2006-02-17 11:53:09

by Bodo Eggert

[permalink] [raw]
Subject: Re: MTD for buffered IO

On Tue, 14 Feb 2006, "?????? ?????? ???????" wrote:

>
>
> > Victor Porton,,, <[email protected]> wrote:
> >
> > > I suggest to add kernel configuration/option to allow to use
> > > an MTD device as a continuation of I/O buffers (for HDD).
> >
> > Why do you want to be limited to MTD devices?
> > What about creating a device-mapper target using JBD instead?
>
> Good idea, which probably already works with current Linux with ext3. But we should be not limited to JDB systems. Your idea works only with JDB, so your idea (despite of being a good idea) is a hack.

NACK, my idea is "porting" the JBD function from ext3 to dm.

> > > One way to implement it would be to add the option for a swap
> > partition/file
> > > to allow to use this swap partition/file as I/O buffer for other device.
> >
> > -EBADIDEA. You'd certainly lose data on power failures.
>
> How it is worse than to lose data on power failures with the standard Linux memory I/O cache?

It will increase the cache and thereby the amount of in-flight data.

> Moreover so we can lose LESS, as Flash memory will preserve swap content unlike RAM.

I didn't asume that you suggest the cache to be preserved, since swap does
not provide the facility. JBD obviously does have this feature, therefore
my suggestion.