Re: advice: 3ware+raid+xfs

[Michael Loftis <mloftis@xxxxxxxxx>]

Depending on your 3Ware card, try to create a RAID with 128K or 256K 
stripes....128K is the 'normal' blocking factor the kernel uses when 
combining reads and writes.  Presuming you're doing mostly sequential 
access a 128K stripe would be optimal.

Now RAID0, vs RAID5 depends....Can you suffer losing a disk and thus losing 
*all* the data?  In RAID0 there is no redundancy, but it is *FAST* very 
fast.  in RAID5 you can lose a drive and run ok well enough to replace the 
drive and rebuild the arry.

RAID5 has some performance penalty, you'll have to look up benchmarks or 
perform your own to find out exactly what the penalty is.

For this much data I would recommend a RAID, and presuming oyu can handle 
the performance drop RAID5, unless the data is replicated elsewhere and 
totally non-critical then go ahead and go RAID0 for some pretty blazing 
fast speeds.

--On Sunday, December 07, 2003 12:14 PM -0500 Gaspar Bakos 
<gbakos@xxxxxxxxxxxxxxx> wrote:

> Dear all,
>
> I am about to build a system for data reduction, but before I do so, I
> thought of posting this in case anyone has useful hints (don't do that!,
> or be careful with...)
>
> It would be a dual xeon mobo + 3ware Escalade card + 4x250Gb (WD) disk,
> running most probably RH9.0 and kernel 2.4.22-xfs. I haven't decided yet
> about the arrangement of the 4x250Gb disks, but definitely there will be
> XFS on them. My possibilities are: (I need total space more than 500Gb)
> 1. JBOD, each disk one partition
> (drawback: I have to take care of not filling either of them)
> 2. RAID-0, one single 1Tb XFS partition
> 3. RAID-5
>
> Any advice would be welcome. I guess people have done this, and
> have experience. E.g.issues related to XFS when one disk
> fails, and has to be replaced. Or mkfs/xfs parameters ideal for mostly
> relatively big files (8Mb and 16Mb) accompanied with very small files
> (<1kB). Recovery issues: I saw xfs_check run out of memory on a single
> 120Gb partition after an unexpected power failure. 3Ware configuration
> issues that might be related to XFS, speed, efficiency.
>
> If the above ide, however, seems viable, just ignore this email.
>
> Being an astronomer, I am not that experienced with sw/hw issues...
> I was  always wondering when people write "we have been testing XFS with
> 60Tb filesystems" (and other magic numbers) - how they do that?
>
> Best wishes
> Gaspar



--
Undocumented Features quote of the moment...
"It's not the one bullet with your name on it that you
have to worry about; it's the twenty thousand-odd rounds
labeled `occupant.'"
  --Murphy's Laws of Combat