[linux-elitists] OpenSolaris

Eugen Leitl eugen@leitl.org
Sat Nov 19 22:26:06 PST 2005


On Sun, Nov 20, 2005 at 01:28:26PM +1100, Jeff Waugh wrote:

> > I still fail to see how a zfs upon a JBOD will magically resurrect my data
> > when a drive fails, as compared to a simple RAID 1. Checksumming, sure,
> > but smartmontools do provide a failure prognosis as well.
> 
> "RAID-Z" does RAID-5 style striping at the logical block level. If you have
> four disks, it stripes across those. Add another, all new transactional
> writes will strip across all 5.

Thanks. Of course I can't put more than 2 SATA drives into the X2100.
(I'm going for two Deathstars T7K250, which is reasonable compromise
in reliability, speed and price), so there is just RAID 1 available
in that constellation. My remote (domestic DSL) backup server is currently a 3-disk
RAID 5 (no hot spare) which I will soon upgrade to 4-disk RAID 5
(one hot spare).

Any suggestions on how to reliably aggregate filesystems, without
sacrificing performance overmuch? I'm currently limited to a private
GBit Ethernet network. My current storage pool in the rack is a DIY dual mini-ITX
blade with 2x300 GB Maxtor (consumer crap, but I got them almost for free)
serving HA NFS over Fast Ethernet (IIRC bonnie++ mumbled something
about 2-3 MBytes/s which is truly pathetic, but sufficient for now).

I can use RAID 1 at machine level, and aggregate that over PVFS. I could
try Lustre or GFS, if it's a pool of Linux machines. VServers seem to
be evolving towards checkpointing and migration capability, so a cluster
approach with diagnostics to route around individual machine failure would
make sense. A global filesystem which could be grown and tolerate loss
at invididual drive and machine level would be good.

Both Lustre and GFS seem to prefer a SAN approach though, no? I don't 
see how I could run ATA over Ethernet, and still have the drives available
in a local file system. OpenAFS could be another option (but not very performant,
apparently).

There are too many options. I'm confused. Any advice from a person in the
trenches? 
 
> (What I haven't figured out is whether it does lazy writes of the old 4 disk
> striped data across the 5 disks, apart from the corruption correction stuff.
> If not, and over time the original 4 disks die, what happens? But for the
> moment, I'll assume something sensible happens... and then ask one of the
> gang.)

-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a>
______________________________________________________________
ICBM: 48.07100, 11.36820            http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://allium.zgp.org/pipermail/linux-elitists/attachments/20051120/cac09f62/attachment.pgp 


More information about the linux-elitists mailing list