[rhelv6-list] You suggestion for 'big' filesystem management Best Practice?

[Prentice Bisbal]

On 10/28/2011 01:32 PM, Peter Ruprecht wrote:
> Greg Swift wrote:
>> On Fri, Oct 28, 2011 at 11:30, Masopust, Christian
>> <christian.masopust at siemens.com
>> <mailto:christian.masopust at siemens.com>> wrote:
>>
>>
>>      > Götz Reinicke wrote:
>>      > > Hi,
>>      > >
>>      > > we plan to set up a big file storage for media files like
>>      > uncompressed
>>      > > movies from student film projects, dvd images etc.
>>      > >
>>      > > It should be some sort of archive and will not bee accessed
>>      > by more than
>>      > > may be 5 people at the same time.
>>      > >
>>      > > The iSCSI RAID we have is about 26TB netto and I'm again
>>      > faced with the
>>      > > question: How many partitions, which filesystem, which
>>      > mount options etc.
>>      > >
>>      > > For the User it would be the most simpel thing, to have one big
>>      > > filesystem she/he could fill with all the data and dont has
>>      > to search
>>      > > e.g. on multiple volumes.
>>      > >
>>      > > On the other hand, if one big filesystem crashes or has do
>>      > be checked it
>>      > > will destroy a lot of data or the check will take hours ...
>>      > >
>>      > >
>>      > > Any suggestions pro or cons are welcome! :-)
>>      > >
>>      > > My favourite for now is 3 to 4 filesystems with the default
>> ext4
>>      > > settings. (Redhat EL 5.7, may be soon 6.1)
>>      > >
>>      > > Thanks and best regards. Götz
>>      >
>>      > If you decide to go with RHEL6, xfs is a good bet for making
>> one big
>>      > filesystem.  We have a setup similar to what you're
>>      > describing and have
>>      > had very solid stability and performance using xfs (default
>>      > filesystem
>>      > and mount settings.)  As far as I can see (and knocking on
>>      > wood), xfs is
>>      > now a lot less flaky than it seemed to be in the past.
>>      >
>>      >   -Peter
>>
>>     I can approve what Peter mentioned. I've been using xfs on my
>>     CentOS 5 system with 2 16TB arrays (each holding one single
>> filesystem)
>>     for several years with absolutely no issues!
>>
>>
>> So in his intial request he mentioned concern about fsck times.  How
>> has this been for you guys (Christian and Peter) ?
>>
>> fwiw, I'm actually mixing both xfs with 30+TB total file system and
>> gluster in a different use case...  I just haven't had to fsck a
>> system yet so I am very curious about how that is performing for others.
>>
>> -greg
>
> In testing, I purposely crashed the system while under light-moderate
> I/O load, and the xfs fs didn't need any recovery when it was
> remounted.  I don't have any real-world experience with how long it
> would take to xfs_check and xfs_repair a fs of that size that had
> gotten corrupted, sorry.  Though I will not be disappointed if I
> manage to avoid gaining that experience!
>
>  -Peter
>

I used to manage a handful of SGI's running IRIX on XFS. If I recall
correctly, an XFS filesystem should NEVER need an fsck or xfs_repair.  
Of course, it's the vendor who says that, not the sys admins working in
the trenches with XFS.  In fact, I don't even think those utilities are
installed as part of the OS - I'm pretty sure I had to install them
manually myself.

That was a few years ago, so my memory might be a bit hazy.

Prentice