ReiserFS on a USB memory stick?

ReiserFS on a USB memory stick?

[Felix E. Klee]

Hi,

I want to run LINUX on a 512MB USB memory stick that supports a minimum
of 100,000 rewrites. I'll mostly be running an EMACS session with
Preview-Latex where I'll be editing and compiling TeX files. Note that I
don't care if the stick fails after 3 months, but it should last about
that long.

Is ReiserFS a suitable file system for that kind of application or are
there better systems? Are there some ways to tune caching so that data
isn't written to the stick as often as it could?

BTW, I already found JFFS2 but it seems to be made for direct access to
flash hardware, while a USB memory stick emulates a hard disk or floppy
super drive.

Felix

-- 
To contact me off list don't reply but send mail to
    felix DOT klee AT inka DOT de
Otherwise your email to me might get automatically deleted!

Re: ReiserFS on a USB memory stick?

[Hans Reiser]

Felix E. Klee wrote:

>Hi,
>
>I want to run LINUX on a 512MB USB memory stick that supports a minimum
>of 100,000 rewrites. I'll mostly be running an EMACS session with
>Preview-Latex where I'll be editing and compiling TeX files. Note that I
>don't care if the stick fails after 3 months, but it should last about
>that long.
>
>Is ReiserFS a suitable file system for that kind of application or are
>there better systems? Are there some ways to tune caching so that data
>isn't written to the stick as often as it could?
>
>BTW, I already found JFFS2 but it seems to be made for direct access to
>flash hardware, while a USB memory stick emulates a hard disk or floppy
>super drive.
>
>Felix
>
>  
>
Compact flash does wear leveling in hardware, I don't know about memory 
sticks.  If people use the FAT filesystem on them, then almost surely  
they do also.;-)

If there is wear leveling, reiserfs will work.  You might want to reduce 
the journal size a bit though.

-- 
Hans

Re: ReiserFS on a USB memory stick?

[Felix E. Klee]

On Fri, 03 Oct 2003 21:06:53 +0400 Hans Reiser wrote:
> >I want to run LINUX on a 512MB USB memory stick that supports a
> >minimum of 100,000 rewrites. I'll mostly be running an EMACS session
> >with Preview-Latex where I'll be editing and compiling TeX files.
> >Note that I don't care if the stick fails after 3 months, but it
> >should last about that long.
> >
> >Is ReiserFS a suitable file system for that kind of application or
> >are there better systems? Are there some ways to tune caching so that
> >data isn't written to the stick as often as it could?
> >
> >BTW, I already found JFFS2 but it seems to be made for direct access
> >to flash hardware, while a USB memory stick emulates a hard disk or
> >floppy super drive.
>
> Compact flash does wear leveling in hardware, 

AFAIK, only the expensive ones do.

> I don't know about
> memory sticks.  If people use the FAT filesystem on them, then almost
> surely  they do also.;-)

I guess not - it's mentioned neither on the packaging nor in the
specifications at
    http://www.twinmos.com/flash/flash_p_mobile_usb20.htm#s


> If there is wear leveling, reiserfs will work.  You might want to
> reduce the journal size a bit though.

I'm a newbie to the internals of file systems so I don't see what would
be the advantage of that other than using less space. Could you tell me
why making the journal size smaller would be an improvement?

BTW, I read on
   http://freshmeat.net/articles/view/212/ 
that the journal isn't cached. Is this really true? If so this would
probably render ReiserFS totally unusable for a flash memory
(even with wear leveling) because there would be far to many accesses.

However, I wonder if other file systems are much better even if they
cache all disk writes. The problem that I see is that there probably are
always parts on the disk that are used for maintainance (e.g. an
inode-table) that get written to over and over again.

But this problem can be somewhat improved to my liking by proper
caching of write accesses. I need a solution where data is written not
more than once to the same sector in less than, say, a minute. Is such a
solution feasible, i.e. are there file systems that cache all write
accesses? Or is this not a question of file systems at all, i.e. is
caching done by LINUX on a layer between the block device and the file
system(if so, this would answer my caching questions above)?

Finally: Do you have an idea which common LINUX file system distributes
writes best (this includes writes to inode-tables, etc.)? You mentioned
FAT before, but this is not really usable for running LINUX on it (other
than with UMSDOS or some such thing).

Felix

-- 
To contact me off list don't reply but send mail to
    felix DOT klee AT inka DOT de
Otherwise your email to me might get automatically deleted!

Re: ReiserFS on a USB memory stick?

[Hans Reiser]

Felix E. Klee wrote:

>
>I'm a newbie to the internals of file systems so I don't see what would
>be the advantage of that other than using less space.
>
none

> Could you tell me
>why making the journal size smaller would be an improvement?
>
>BTW, I read on
>   http://freshmeat.net/articles/view/212/ 
>that the journal isn't cached. Is this really true?
>
Not at all.  All journaling file systems cache the journal, with the 
possible exception of NTFS whose performance sucks.

Also, preserve lists were replaced in ReiserFS by the journaling code 
many years ago.  Maybe I should work on updating our website today, and 
displace all the V3 docs with V4 docs....

In my benchmarks at least, ext3 performs better than XFS except for very 
large files and large directories.

> If so this would
>probably render ReiserFS totally unusable for a flash memory
>(even with wear leveling) because there would be far to many accesses.
>
>However, I wonder if other file systems are much better even if they
>cache all disk writes. The problem that I see is that there probably are
>always parts on the disk that are used for maintainance (e.g. an
>inode-table) that get written to over and over again.
>
>But this problem can be somewhat improved to my liking by proper
>caching of write accesses. I need a solution where data is written not
>more than once to the same sector in less than, say, a minute. Is such a
>solution feasible, i.e. are there file systems that cache all write
>accesses? Or is this not a question of file systems at all, i.e. is
>caching done by LINUX on a layer between the block device and the file
>system(if so, this would answer my caching questions above)?
>
>Finally: Do you have an idea which common LINUX file system distributes
>writes best (this includes writes to inode-tables, etc.)? You mentioned
>FAT before, but this is not really usable for running LINUX on it (other
>than with UMSDOS or some such thing).
>

No no, FAT DOES NOT distribute writes, which is why if FAT is used on 
something you know it must use wear leveling (or it is a piece of trash).

I suggest you simply ask your manufacturer if your stick does wear 
leveling.  It most likely does.

>
>Felix
>
>  
>


-- 
Hans

Re: ReiserFS on a USB memory stick?

[Hans Reiser]

I should also add, in regards to the article, if you are using XFS on 
Linux, it is not as stable as ext3 and reiserfs.  I think the article 
author doesn't understand just how much code gets changed during a 
port.  It took them years to do the port.  ReiserFS is more stable 
mostly because it was finished before the others (we worked pretty hard 
to make that happen).

-- 
Hans

Re: ReiserFS on a USB memory stick?

[Felix E. Klee]

On Sat, 04 Oct 2003 08:50:22 +0400 Hans Reiser wrote:
> >Finally: Do you have an idea which common LINUX file system
> >distributes writes best (this includes writes to inode-tables, etc.)?
> >You mentioned FAT before, but this is not really usable for running
> >LINUX on it (other than with UMSDOS or some such thing).
> 
> No no, FAT DOES NOT distribute writes, which is why if FAT is used on 
> something you know it must use wear leveling (or it is a piece of
> trash).

Ah, I misunderstood your reply to my original message. I thought you
were referring to the fact that files in a FAT based system get
fragmented fast and therefore data gets distributed (which actually
doesn't matter since the file allocation table is probably the
most problematic part).

> I suggest you simply ask your manufacturer if your stick does wear 
> leveling.  It most likely does.

I just got answers from TwinMOS international and German technical
support. One said that all their sticks use wear leveling technology,
the other said, more specifically, that it uses so called "random write"
technology. 

So, I guess it doesn't really matter what file system I use if I want to
minimize the wear on the stick. 

BTW, as I'm running LINUX on the stick, it might be a good idea to
backup, erase and restore the entire data on the stick once in a while.
Due to wear leveling technology this might interchange the physical
position of areas with rewritable data (/home, /var, etc.) and areas
with read-only data (/usr, etc.). This might also be a good chance to
try out different file systems ;)

Felix


-- 
To contact me off list don't reply but send mail to
    felix DOT klee AT inka DOT de
Otherwise your email to me might get automatically deleted!