suggestion on use of linux-mtd

suggestion on use of linux-mtd

[Andre]

My apologies for my previous post
http://lists.infradead.org/pipermail/linux-mtd/2005-June/012816.html

I managed to find the mtd-jffs2-howto in the mtd source tree, which contains
all the instructions for building the kernel.

I have been running ext3 on diskonchip2000 (using m-sys trueffs driver) for
quite a while now and have not run into any problems yet, although I never
did any asynchronous power off tests. I can boot off diskonchip using a
special version of lilo and initrd.

I would really like to try linux-mtd, however I am not quite sure if it is
suitable for my use of the diskonchip. First of all, my application will
create up to 32KB/s of data, and it will run for several hours (if not days
at lower data rate). The diskonchip will be at least 384MB up to a maximum
of 1GB. The OS (trimmed down LFS5.0, patched 2.4.27) is about 35M and lives
in the same partition. Asynchronous power offs cannot be ruled out due to
the nature of the instrument that will host the cpu board with the
diskonchip.

The mtd-jffs-howto refers to NFTL for diskonchip devices, but also talks
about JFFS2 and how production worthy it is and it looks like some work has
been done in the area of powercycle testing, but the link provided is not
valid.

What exactly does one get with *just* NFTL, i.e. does it makes sense to run
ext3 on top of NFTL? Is NFTL production ready?

There is also a comment that says JFFS2 does not work on DOC, but I assume
that comment is old. What exactly is the status of JFFS2 for diskonchip
devices?

regards,
AWS

Re: suggestion on use of linux-mtd

[Coywolf Qi Hunt]

On 6/14/05, Andre <andre@rocklandocean.com> wrote:
> My apologies for my previous post
> http://lists.infradead.org/pipermail/linux-mtd/2005-June/012816.html
> 
> I managed to find the mtd-jffs2-howto in the mtd source tree, which contains
> all the instructions for building the kernel.
> 
> I have been running ext3 on diskonchip2000 (using m-sys trueffs driver) for
> quite a while now and have not run into any problems yet, although I never
> did any asynchronous power off tests. I can boot off diskonchip using a
> special version of lilo and initrd.
> 
> I would really like to try linux-mtd, however I am not quite sure if it is
> suitable for my use of the diskonchip. First of all, my application will
> create up to 32KB/s of data, and it will run for several hours (if not days
> at lower data rate). The diskonchip will be at least 384MB up to a maximum
> of 1GB. The OS (trimmed down LFS5.0, patched 2.4.27) is about 35M and lives
> in the same partition. Asynchronous power offs cannot be ruled out due to
> the nature of the instrument that will host the cpu board with the
> diskonchip.
> 
> The mtd-jffs-howto refers to NFTL for diskonchip devices, but also talks
> about JFFS2 and how production worthy it is and it looks like some work has
> been done in the area of powercycle testing, but the link provided is not
> valid.
> 
> What exactly does one get with *just* NFTL, i.e. does it makes sense to run
> ext3 on top of NFTL? Is NFTL production ready?

No. NFTL is for compatibility purpose.

> 
> There is also a comment that says JFFS2 does not work on DOC, but I assume
> that comment is old. What exactly is the status of JFFS2 for diskonchip

-- 
Coywolf Qi Hunt
http://ahbl.org/~coywolf/

Re: suggestion on use of linux-mtd

[Raphael Zimmerer]

On Mon, Jun 13, 2005 at 05:04:46PM -0700, Andre wrote:
> There is also a comment that says JFFS2 does not work on DOC, but I assume
> that comment is old. What exactly is the status of JFFS2 for diskonchip
> devices?

Depends on what driver you are using. With the drivers based on the nand
subsystem there shouldn't be any problem. The driver is named
diskonchip.c.

- kille

Re: suggestion on use of linux-mtd

[Sergei Sharonov]

Hi,

> I have been running ext3 on diskonchip2000 (using m-sys trueffs driver) for
> quite a while now and have not run into any problems yet, although I never
> did any asynchronous power off tests. 

I am also interested in power fail tolerance of ext3 but could not find
any test data. People mentioned that it should be better then other 
journaling filesystem (such as reserfs, jfs, xfs) because of different
journaling scheme. 

> The diskonchip will be at least 384MB up to a maximum
> of 1GB. 

You may forget about jffs2 on such large flashes. Depending on the size of the
writes that filled the disk it will take anywhere from ten minutes to hours
to mount jffs2. The only other alternative that uses mtd is yaffs. It does
mount significantly faster, but the user base is small and there is no active
development going on (as far as I can tell).

> in the same partition. Asynchronous power offs cannot be ruled out due to
> the nature of the instrument that will host the cpu board with the
> diskonchip.

I know people that did extensive power-cycling testing on jffs2 with very
good results. I tried cycling yaffs and it fails on me about after 100-200
cycles.

> What exactly does one get with *just* NFTL, i.e. does it makes sense to run
> ext3 on top of NFTL? Is NFTL production ready?

Are you going to power cycle your nftl+ext3 setup? I would be
interested in results.

Best regards,

Sergei Sharonov

Re: suggestion on use of linux-mtd

[Artem B. Bityuckiy]

Sergei Sharonov wrote:
> I am also interested in power fail tolerance of ext3 but could not find
> any test data. People mentioned that it should be better then other 
> journaling filesystem (such as reserfs, jfs, xfs) because of different
> journaling scheme. 
> 
Hi Sergey,

BTW,

I remember you were needing atomicity. Reiser4 supports that.
It supports start transaction/commit semantic, although there is no
such syscall in Linux (yet). Nevertheless, Reiser4 is not stable yet.

-- 
Best Regards,
Artem B. Bityuckiy,
St.-Petersburg, Russia.

Re: suggestion on use of linux-mtd

[Andre]

Sergei Sharonov <sergei.sharonov <at> halliburton.com> writes:

> You may forget about jffs2 on such large flashes. Depending on the size of the
> writes that filled the disk it will take anywhere from ten minutes to hours
> to mount jffs2. The only other alternative that uses mtd is yaffs. It does
> mount significantly faster, but the user base is small and there is no active
> development going on (as far as I can tell).

ok, I will not pursue jffs2.

> > What exactly does one get with *just* NFTL, i.e. does it makes sense to run
> > ext3 on top of NFTL? Is NFTL production ready?
> 
> Are you going to power cycle your nftl+ext3 setup? I would be
> interested in results.

Yes, as soon as I find out the state of NFTL. I got a little scared by the 
following comment in the howto:

quote: "DO NOT EVER USE THE nftl_format UTILITY WITHOUT FIRST SEEKING ADVICE
ON THE MAILING LIST. It will erase all blocks on the device,
potentially losing the factory-programmed information about bad
blocks. (Someone really ought to fix it one of these days - ed)"

Does NFTL include wear leveling and/or bad block remapping?