HP ML350e G8 Raid Controller P222, 4x SAS 1TB Raid 5, DVD SLES SP1

I suspect it's because you're presenting it with a logical disk greater than 2 TiB (2.2TB). Use intelligent provisioning to access the Array Configuration Utility and delete the current logical disk, then create a small one for boot and the OS. You can create a second large logical disk later for data.

More current linux kernels have the necessary fixes in fdisk & diskpart to prevent such things from happening in the first place. For what it is worth, the limit is FFFFFFFD 512-byte blocks, so you can use your handy-dandy hex to decimal calculator to convert that to bytes.    Anything larger won't work unless you have an Itanium CPU or use a UNIX variant instead of LINUX.

> (AMD64/EM64T).
That's the AMD driver but the server is Intel based.
Firmware for the controller won't harm but shouldn't be needed.

There isn't an expander on the backplane for that machine, HP's cross-reference table must be up the spout again.

Are you trying to use Intelligent Provisioning to do the install? There's specific mention of your OS in the release notes vis

 Linux-specific issues
• If SUSE Linux Enterprise Server 11 SP1 is updated using the base operating system image in Intelligent Provisioning, the connected operating system media will be scanned as Valid media, but after booting connected media hard disk will not be detected.
Suggested action: When installing SUSE Linux Enterprise Server 11 SP1 via Intelligent Provisioning, a kISO (kernel update ISO) must be used.

Don't know what that means but it looks the same as your issue.

This says it's the chipset causing it, but presumably SP2 is better than using the KISO anyway although it would be a smaller download.

http://h20566.www2.hp.com/portal/site/hpsc/template.PAGE/public/kb/docDisplay/?spf_p.tpst=kbDocDisplay&spf_p.prp_kbDocDisplay=wsrp-navigationalState%3DdocId%253Demr_na-c03237878-1%257CdocLocale%253D%257CcalledBy%253D&javax.portlet.begCacheTok=com.vignette.cachetoken&javax.portlet.endCacheTok=com.vignette.cachetoken

Certainly easier for maintenance to have a separate pair of OS disks but you can still use the space that's left over in the current setup just by creating another logical disk on it and treating it as a second disk.

No, use the md driver to do host-based RAID1.  heck you can even configure it to do 4-way mirroring of the /boot partition and then go RAID5 with another slice of those same disks, or something in the middle.

Personally, if this was my system I would go with md0 = RAID1 of 50GB on two disks (/boot & /) , md1= RAID1 of 50GBon two more (the 2nd one can be /tmp & swap, then use the remaining 950GB of the 4 disks to be a large RAID10 for everything else.

No, i would let the O/S do the RAID with the md driver.  Configure the controller to present 4 disk drives. Reason is that the O/S will do a better job of distributing the I/O load, and it certainly has more caching capability.  RAID1/10 has effectively no overhead and LINUX RAID will do read load balancing (as the HP will).

But the HP RAID will not balance the I/O as well because the O/S is privy to knowing how you use your data. Plus it can set block sizes to match the I/O of the md driver.   You're going to be doing 4KB I/O at the disk in some partitions, 64KB in others, you have that flexibility with the MD driver.  But with the HP you have all I/Os to the disk same block size, perhaps 64KB whether you need to write that much data or not.

The controller (well the driver really) is privy to the data and will re-order it in conjunction with the controller. You do not have to have all I/Os to the disk at the same size since you can create several logical disks with different RAID levels and strip sizes. If you meant all I/Os to the logical disk have to be the same size then that's correct.

I would leave it as you have it, disabling RAID on a Smart Array controller isn't possible on a ProLiant (well it is but you would have to temporarily put into an Itanium based Integrity server to do that) and it would be a bit messy having 4 RAID 0s.

RAID 1+0 is RAID 10 at least under the terminology currently used by HP. The old Compaq controllers/ACU used to call it RAID 0+1 erroneously but it was still RAID 10 really.

RAID 0+1, 10, 1+0 have all hopelessly been munged by marketeers.  They are effectively the same in the sense that all do striping and mirroring together.

There is absolutely no difference between RAID 10 and RAID 1+0 under the most common usage of the terms. You are probably thinking of what most people call RAID 0+1. Even when Compaq (erroneously) offered 0+1 in the ACU and documentation their algorithm was actually RAID 10.

Under the common usage of the terminology RAID 1+0 is a stripe of mirrors which is what RAID 10 is defined as.

What is less reliable is what is most commonly called RAID 0+1 where you stripe before you mirror. No server RAID controller I know uses that algorithm, the only hardware controller I know of that could use 0+1 was the HSG80 which could do both. The only software I know that does 0+1 is Solaris Volume Manager which can also do both but will automatically treat a volume setup as 0+1 as 1+0 if disk sizes allow.

http://docs.oracle.com/cd/E19683-01/806-6111/about-mirrors-7/index.html explains the difference and how Solaris Volume Manager works around the 0+1 reliability problem if it can.


As far as hot spares are concerned one just burned my customer's data vis: 3 disk RAID 5 plus spare, one predictive failure and one disk that kept going offline. Reboot and the controller decided to take the predictive failure disk offline and rebuild onto the spare. Disk that kept going offline went offline so volume died whereas it would still be working on one good disk and one predictive failure. It didn't work with one good disk and one half-built spare.

No, UtahTN .. you are under the assumption that the only possible fault is a catastrophic drive failure.   All it takes is just one unreadable block on one of the surviving drives in a RAID1,10,5 situation and you have partial data loss.  (Even if only one disk failed).

RAID6 is only mechanism from choice above that has 100% of the data protected by double redundancy.

I have no idea what you should do.  RAID6 has a performance penalty, but the flip side is that it has advantages in data integrity & availability.  There is no general right answer.  

Ask yourself what it costs if you lose data and have to restore from a backup vs. a 0-25% performance hit over RAID5 (depending on a lot of variables)

I'd stick to RAID 10 in most cases with that controller, it does a background surface analysis when the disks are otherwise idle so unless it's busy 24x7 it's unlikely you'll suffer from a failed disk plus bad blocks although admittedly they can develop at any time.

Maybe a better analogy is a simple one. At some point a HDD will fail (you have 100% probability).  So when this happens, will you be sweating and hoping that you don't lose a 2nd drive and all of your data, and have to restore from backup?

Or do you want to be able to tell whomever calls you that you'll worry about it Monday or whenever you get back from vacation and that they can still lose one more drive before they are exposed for data loss.

If you're that worried about it then buy more disks and use 3-way mirroring which gives even greater data protection without the performance loss. Both RAID 6 and Advanced Data Mirroring require the Advanced Pack license.