ReiserFS on a flash device?

ReiserFS on a flash device?

[Steve Snyder]

I was taken aback by Yury Umanets' recent suggestion 
(http://groups.google.com/groups?hl=en&lr=lang_en&ie=UTF-8&safe=off&th=d50aa0bf6c7f26d0&seekm=1059231274.28094.40.camel%40haron.namesys.com.lucky.linux.kernel#link11) 
that ReiserFS is not an appropriate for a flash device.

I am currently using it (via Linux kernel v2.4.21) on an E-Disk 
(http://www.bitmicro.com/products_edisk_35_scsiw.php) flash device as the 
front-end of a low-traffic Squid Web cache.  The Squid experts recommend 
the use of ReiserFS (with the notail option) for it's superior 
performance in handling lots of small files.  I understand that flash has 
a finite number of write cycles, but for this application performance is 
critical, and disk I/O (access times, not STR) is the bottleneck in Squid 
performance.

A quote from the post referred to above:

>But, if you are still want to use reiserfs for flash device, you should
>do at least the following: 
>
>(1) Make the journal substantial smaller of size.
>(2) Don't turn tails off. This is useful to prolong flash live.

More detail, please?  How is journal size a factor?  How does notails 
reduce flash longevity?

Thanks.

Re: ReiserFS on a flash device?

[Shawn Rutledge]

I wanted to use ReiserFS for an embedded system, but on a 32 or 64 meg CF
card, v3 takes up too much of it for a journal, so it wasn't practical.
Is there a way to make the journal much smaller, or have it grow
on-demand rather than being fixed-size?  In practice the CF card won't 
be written to very much (for longevity reasons), but I don't ever want 
an embedded system to waste time fscking itself and prompting what to 
do about any errors; and at the same time don't see why it should be 
required to be a read-only filesystem, as is often done, because it makes
the occasional upgrade harder.  And apparently JFFS2 is only for simple 
memory-mapped FLASH chips without built-in write-balancing like CF cards 
have.

I use the mainstream ReiserFS in Linus' kernels on all my home machines
and love it.  It's so fast and stable as a rock, and the only problems 
I've had were hardware-related.

On Sat, Jul 26, 2003 at 04:40:13PM -0700, Steve Snyder wrote:
> I was taken aback by Yury Umanets' recent suggestion 
> (http://groups.google.com/groups?hl=en&lr=lang_en&ie=UTF-8&safe=off&th=d50aa0bf6c7f26d0&seekm=1059231274.28094.40.camel%40haron.namesys.com.lucky.linux.kernel#link11) 
> that ReiserFS is not an appropriate for a flash device.
> 
> I am currently using it (via Linux kernel v2.4.21) on an E-Disk 
> (http://www.bitmicro.com/products_edisk_35_scsiw.php) flash device as the 
> front-end of a low-traffic Squid Web cache.  The Squid experts recommend 
> the use of ReiserFS (with the notail option) for it's superior 
> performance in handling lots of small files.  I understand that flash has 
> a finite number of write cycles, but for this application performance is 
> critical, and disk I/O (access times, not STR) is the bottleneck in Squid 
> performance.
> 
> A quote from the post referred to above:
> 
> >But, if you are still want to use reiserfs for flash device, you should
> >do at least the following: 
> >
> >(1) Make the journal substantial smaller of size.
> >(2) Don't turn tails off. This is useful to prolong flash live.
> 
> More detail, please?  How is journal size a factor?  How does notails 
> reduce flash longevity?
> 
> Thanks.
> 
> 

-- 
  _______                                                Shawn T. Rutledge
 (_  | |_)                      e_cloud@yahoo.com * http://ecloud.org:8080
 __) | | \________________________________________________________________

Re: ReiserFS on a flash device?

[Yury Umanets]

On Sun, 2003-07-27 at 03:40, Steve Snyder wrote:
> I was taken aback by Yury Umanets' recent suggestion 
> (http://groups.google.com/groups?hl=en&lr=lang_en&ie=UTF-8&safe=off&th=d50aa0bf6c7f26d0&seekm=1059231274.28094.40.camel%40haron.namesys.com.lucky.linux.kernel#link11) 
> that ReiserFS is not an appropriate for a flash device.
> 
> I am currently using it (via Linux kernel v2.4.21) on an E-Disk 
> (http://www.bitmicro.com/products_edisk_35_scsiw.php) flash device as the 
> front-end of a low-traffic Squid Web cache. 

>  The Squid experts recommend 
> the use of ReiserFS (with the notail option) for it's superior 
> performance in handling lots of small files.
Yes, but they probably did not mean use it with flash device? 

Actually I have took a look onto edisk specifications. It has SCSI
interface. System will not be concerned about its physical details. Also
it has lots of data security features, and probably you should not be
worry about its exhausting :)

>   I understand that flash has 
> a finite number of write cycles, but for this application performance is 
> critical, and disk I/O (access times, not STR) is the bottleneck in Squid 
> performance.

> 
> A quote from the post referred to above:
> 
> >But, if you are still want to use reiserfs for flash device, you should
> >do at least the following: 
> >
> >(1) Make the journal substantial smaller of size.
> >(2) Don't turn tails off. This is useful to prolong flash live.
> 
> More detail, please?  How is journal size a factor?  

Default journal size is too big (32MB) for usual embedded device. I
mean, that flash is expensive and it is pity to loss 32MB for journal :)

> How does notails 
> reduce flash longevity?

Tail policy "tails turned on" causes do not use redundant block (4K) for
small piece of data, as it may be saved in tree. This may be considered
as compression with very bad rate :)

> 
> Thanks.
> 
> 
-- 
We're flying high, we're watching the world passes by...

Re: ReiserFS on a flash device?

[Yury Umanets]

On Sun, 2003-07-27 at 04:03, Shawn Rutledge wrote:
> I wanted to use ReiserFS for an embedded system, but on a 32 or 64 meg CF
> card, v3 takes up too much of it for a journal, so it wasn't practical.

> Is there a way to make the journal much smaller, or have it grow
> on-demand rather than being fixed-size? 

You can use smaller journal. Say 512 blocks (2MB). And in this case you
need patches for supporting non-standard journal. Or use last kernel pre
or wait next kernel release.

If you want boot from that flash with non-standard reiserfs on it, use
GRUB patch from our website for supporting non-standard journal.


>  In practice the CF card won't 
> be written to very much (for longevity reasons), but I don't ever want 
> an embedded system to waste time fscking itself and prompting what to 
> do about any errors; and at the same time don't see why it should be 
> required to be a read-only filesystem, as is often done, because it makes
> the occasional upgrade harder.


>   And apparently JFFS2 is only for simple 
> memory-mapped FLASH chips without built-in write-balancing like CF cards 
> have.

As I know, JFFS2 uses compression for storing data. And if so, driver 
cannot map flash data to memory.

> 
> I use the mainstream ReiserFS in Linus' kernels on all my home machines
> and love it.  It's so fast and stable as a rock, and the only problems 
> I've had were hardware-related.
> 
> On Sat, Jul 26, 2003 at 04:40:13PM -0700, Steve Snyder wrote:
> > I was taken aback by Yury Umanets' recent suggestion 
> > (http://groups.google.com/groups?hl=en&lr=lang_en&ie=UTF-8&safe=off&th=d50aa0bf6c7f26d0&seekm=1059231274.28094.40.camel%40haron.namesys.com.lucky.linux.kernel#link11) 
> > that ReiserFS is not an appropriate for a flash device.
> > 
> > I am currently using it (via Linux kernel v2.4.21) on an E-Disk 
> > (http://www.bitmicro.com/products_edisk_35_scsiw.php) flash device as the 
> > front-end of a low-traffic Squid Web cache.  The Squid experts recommend 
> > the use of ReiserFS (with the notail option) for it's superior 
> > performance in handling lots of small files.  I understand that flash has 
> > a finite number of write cycles, but for this application performance is 
> > critical, and disk I/O (access times, not STR) is the bottleneck in Squid 
> > performance.
> > 
> > A quote from the post referred to above:
> > 
> > >But, if you are still want to use reiserfs for flash device, you should
> > >do at least the following: 
> > >
> > >(1) Make the journal substantial smaller of size.
> > >(2) Don't turn tails off. This is useful to prolong flash live.
> > 
> > More detail, please?  How is journal size a factor?  How does notails 
> > reduce flash longevity?
> > 
> > Thanks.
> > 
> > 
-- 
We're flying high, we're watching the world passes by...

Re: ReiserFS on a flash device?

[Mike Fedyk]

On Sun, Jul 27, 2003 at 02:55:31PM +0400, Yury Umanets wrote:
> Default journal size is too big (32MB) for usual embedded device. I
> mean, that flash is expensive and it is pity to loss 32MB for journal :)

What?  You mean it isn't reduced in size based on block device size like
with ext3?

IIRC, it defaults to 2-4MB on a 100MB device.

Re: ReiserFS on a flash device?

[Yury Umanets]

On Mon, 2003-07-28 at 06:33, Mike Fedyk wrote:
> On Sun, Jul 27, 2003 at 02:55:31PM +0400, Yury Umanets wrote:
> > Default journal size is too big (32MB) for usual embedded device. I
> > mean, that flash is expensive and it is pity to loss 32MB for journal :)
> 
> What?  You mean it isn't reduced in size based on block device size like
> with ext3?

Yes, is isn't. Use reiserfstune to make it smaller.
> 
> IIRC, it defaults to 2-4MB on a 100MB device.
-- 
We're flying high, we're watching the world passes by...

Re: ReiserFS on a flash device?

[Shawn Rutledge]

On Mon, Jul 28, 2003 at 08:53:57AM +0400, Yury Umanets wrote:
> On Mon, 2003-07-28 at 06:33, Mike Fedyk wrote:
> > On Sun, Jul 27, 2003 at 02:55:31PM +0400, Yury Umanets wrote:
> > > Default journal size is too big (32MB) for usual embedded device. I
> > > mean, that flash is expensive and it is pity to loss 32MB for journal :)
> > 
> > What?  You mean it isn't reduced in size based on block device size like
> > with ext3?
> 
> Yes, is isn't. Use reiserfstune to make it smaller.

Well I'm glad to hear it's possible to use ReiserFS on small devices,
but I think that requiring a kernel patch, or a GRUB patch, to handle
different journal sizes is not good.  Are there plans to have the kernel
and GRUB detect it at runtime?

-- 
  _______                                                Shawn T. Rutledge
 (_  | |_)                      e_cloud@yahoo.com * http://ecloud.org:8080
 __) | | \________________________________________________________________

Re: ReiserFS on a flash device?

[Yury Umanets]

On Mon, 2003-07-28 at 11:05, Shawn Rutledge wrote:
> On Mon, Jul 28, 2003 at 08:53:57AM +0400, Yury Umanets wrote:
> > On Mon, 2003-07-28 at 06:33, Mike Fedyk wrote:
> > > On Sun, Jul 27, 2003 at 02:55:31PM +0400, Yury Umanets wrote:
> > > > Default journal size is too big (32MB) for usual embedded device. I
> > > > mean, that flash is expensive and it is pity to loss 32MB for journal :)
> > > 
> > > What?  You mean it isn't reduced in size based on block device size like
> > > with ext3?
> > 
> > Yes, is isn't. Use reiserfstune to make it smaller.
> 

> Well I'm glad to hear it's possible to use ReiserFS on small devices,
> but I think that requiring a kernel patch, or a GRUB patch, to handle
> different journal sizes is not good.  Are there plans to have the kernel
> and GRUB detect it at runtime?
Support for non-standard journal is merged to kernel recently. Use last
pre or wait until next stable kernel is released.

GRUB is able to with with it too. Use
ftp://ftp.namesys.com/pub/misc-patches/grub-0.93-reiserfs-non-standard-journal.diff for that.

Regards.

-- 
We're flying high, we're watching the world passes by...