[Wolves] backing up methods

Adam Sweet adam at adamsweet.org
Fri May 8 13:17:01 UTC 2009


DragonMaster wrote:
>> Ah, hang on...  Is it a USB hard drive or a USB flash drive?  Sorry I
>> can't remember.  Though I think you said it's a 1TB drive so it's got
>> to be a hard drive.  It's alright, the alrm bells started to clamour
>> as ext3 being a journalling system would kill a flash drive in not too
>> long a period.  It's fine on a proper whirling platters hard drive
>> though :-)
>> --
> 
> Really? why's that - surely the journal only gets updated on a write, so
> you would get a few more writes than on an ext2 partition, but nothing
> disastrous?

Well the journal would be updated for every write and I believe the
location of the journal is always in the same place on a disk, which
means that every write to the filesystem means a write to the same
groups of cells occupied by the journal. The default ext3 filesystem
behaviour is to update a file's access time on every read, which means a
journal write, so every time you read or write a file, you write to the
journal.

A growing list of distributions are disabling this nowadays (at least
Ubuntu anyway). You can use the noatime option to disable file access
time updates entirely or relatime to only update the access time when it
is older than the file's modified or creation time, as some applications
need a sensible access time to work properly.

That said, I think, according to talk given by Val Henson (Linux kernel
filesystems and ZFS developer) at LugRadio Live in San Francisco last
year, flash storage is a lot more reliable than it used to be. Allegedly
around 5% of the cells will die after the around 100,000 writes while
the rest can survive between 1 and 5 million.

Journalling and logging operates on the same files and disk areas
causing certain groups of cells to die quicker than the rest of the
disk, so I have heard of some netbook users using ext2 (ie no
journalling) with noatime and setting up some of their RAM as a RAM
drive and mounting /var/log on it. Also manufacturers have started using
wear-leveling:

http://en.wikipedia.org/wiki/Wear_leveling

Though Val Henson pointed out that quite often, wear leveling isn't
written by filesystem specialists and is executed on lowest common
denominator chips on the disk itself. Not a biggie for home users I
don't suppose but not something you would want to run a business on.

Val Henson talk:

http://video.google.com/videoplay?docid=-713335709875581296&hl=en or
http://www.archive.org/details/LRL_USA_2008_Storage_for_startups

Ad



More information about the Wolves mailing list