[Voyage-linux] LEDs, APT repository, MySQL in RAM questions

[Jordi Puigsegur]

Andreas,

As I understand, the CF cards already implement wear leveling algorithms.
Each time a block is updated / written a new block in the CF is used.

Jordi.

2010/7/13 Andreas Delleske <delleske at vauban.de>

> Dear Jordi,
>
> thank you for your friendly reply!
>
> > We use adjtimex + ntpd in order to have a ALIX3D2 synchronized. I
> > wrote how we configured it in
> >
> http://code.google.com/p/wfrog/wiki/InstallOn_Alix3D2_voyageLinux0_6_2#Clock_sync
>
> Thanks, I've found that page already.. :-)
>
> I found out that I could install adjtimex package as soon as I changed
> the apt-sources to the german server..
>
> Now the only buggy package that I need is ssmtp (smtp forwarder to a
> smarthost), but i works when ignoring errors when installing.
>
> > I set up a read-write partition for the database and write a record
> > each 5 min. So far the ALIX 3 D 2 with a 1Gb SLC Compact Flash has
> > been working for half a year perfectly. We'll have to wait a couple of
> > years to know how long the CF last!!!
>
> Yes..
>
> If we think about wear leveling, there is a lot of options:
>
> Here, we will have only tens to hundreds of write updates per day, not
> thousands, and I have 4 GByte Flash memory, only 400 MB of which are
> used so far. The part where the changes appear (the database) is only
> 3 MB now and will 10 MB in some years.
>
> If we assume that every change rewrites 100 kByte (much less in the
> database's data, but with every update one or three flash blocks have
> to be rewritten..) we rewrite max. 10 MB per day = 3,65 GBytes per
> year - within one year, the free space on that disk is rewritten only
> once (statistically)  that is, if we have a proper wear-leveling
> algorithm.  If every flash block can be rewritten only 10.000 times,
> can we think of 10.000 years? I think that would be enough. :-) Or did
> I go wrong?
>
> Does th CF card "know" which blocks are free and use them
> automatically with a round-robin-method? If the CF knows nothing about
> filesystems, it would be much less: Assume we write always in the same
> 10 MB, the CF will last 10.000 days = 27,3 years. Shouldnt even that
> be enough?
>
> I see that if we have one small table in the DB which gets rewritten
> always in the same spot, things get worse: 10.000 operations would be
> accomplished after only a hundred days... maybe it is crucial to put
> only that table in RAM and let the rest of the database grow slowly
> without any concern?
>
> However, it would be great to be able to check the "sanity" of the
> CF-disk like with the smartmontools and in some years just swap the
> disk (they are inexpensive anyway).
>
> Or we could move that table from time to time to another location on
> the flash ("Manual" wear-leveling)? Lets say we create a copy of the
> full database every day and leave to old copy unattended until the
> free space is filled up so with every update a new spot is used? Then,
> with 3,5 GByte free space divided by 10 MByte a day we'd fill the disk
> within a year - but could we redo that 10.000 times? That would be
> enough as well.
>
> --
> Kind regards
> Andreas
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.voyage.hk/pipermail/voyage-linux/attachments/20100715/06cec46b/attachment-0001.html>