Raid 5 + 1 (51)

Yes, we call that RAID 50.

http://support.dell.com/support/edocs/storage/RAID/

That is where you can find the user's guides for all the RAID controllers we offer.

Mark,

   Thank you, the documentation says you are correct. In that case the only way to get a RAID 51 on the Dell SCSI RAID hardware is to create two identical RAID 5's then mirror them in the operating system. (See Dynamic volumes in the Array manager documentation) Dynamic volumes = bad, but that's my personal opinion.

Hello - I'd just like a word of added confirmation here. I just read in your 'raid primer' (http://www.dell.com/downloads/global/products/pvaul/en/RAIDPrimerWP.pdf) your descriptions of all the possible configurations - yet there I see that the RAID 5/0 diagram actually describes a controller-mirrored (duplexed) RAID 5 array (no added striping described in the '0' level twinning both arrays) - this is confusing to say the least. Is your RAID 5/0 equal to what most would describe as RAID 5/1 (http://www.pcguide.com/ref/hdd/perf/raid/levels/multLevel15-c.html) (or duplexed RAID 5)?

I have four Dell PE1950's, each with a perc6/E sata contoller connected to two md1000's (donc eight in total). 

As an alternate soution, I am thinking to the architecture and whether to cluster two PE1950's (so four md1000's) using an application like 'heartbeat' - this would seem most fail-safe for both servers, sata controllers and raid5 arrays.

Thank you very much in advance. 

"Lsi based adapters (Perc) implementing raid 50 are mirrored as in the article's definition of raid 51."

Whew, thank you for your detailed reply. I'll have to look up the above though (lsi based adaptors?) as I'm rather new to nested raids and their methodology. If the explanation is simple, please do elaborate.

I do like your raid 10 idea from a security point of view, but unfortunately data duplication (aka backup) must be simultanious as our service must be available around the clock with no data loss at all (difference between 'restore' disk and down disk).

Thanks again, best. 

"As an alternate solution, I am thinking to the architecture and whether to cluster two PE1950's"

If you want headaches, complexity and lower performance, cluster . 

Personally, if I had the resources, as you do, I would create a raid 10, with more than 4 disks ( the higher disk count in raid 10 lower the possible array failure rate), with hotspare(s). This gives really good performance, unlike raid 50 (due to double parity creation) and is close to the safety of raid 50. You could automate a nightly  image, or update a cloned disk on a separate disk set for added safety, with the likes of Acronis. With a update clone, should the raid 10 array fail, which is extremely unlikely, you could set the machine to boot to the cloned disk set.

Lsi based adapters (Perc) implementing raid 50 are mirrored as in the article's definition of raid 51.

Correction as to the above line, Mark_A_Smith is correct, in raid 50, the individual arrays are spanned not mirrored.

Ps. Don't believe everything you read....as in the above correction and  in the paragraph the author refers to raid 01, he is actually referring to raid 10, raid 10 is safer then raid 01, raid 10 can lose up to half the drive array members of an array before failure, raid 01 can only lose 1 disk .

After much more reading, I am seriously considering your RAID 10 proposition. The architecture vis-à-vis the perc6/E controller could be tricky though - for higher safety I could make RAID 1 pairs between two MD1000's, then apply the 0 level to these - with a few hotspares (global) ready to take over in case of failure.

One thing is bugging me with this configuration though - the possibility of a perc6/E failure, or the failure of a server itself. It is for this that I was considering clustering servers, each managing its own 'first level' RAID array (the clustering serving the role of a duplex or mirror (or second RAID 1) level), but implementing this seems daunting. I'm not even sure if it is possible to cluster two servers *and* their respective external storage RAID arrays.

You can cluster a few ways. Cluster involving just sharing the disk array, I have never done as I consider it foolish, as the array is the weak point. You can cluster complete servers, which as you point out is expensive/complex, and not failsafe unless it is tested/monitored carefully. To be truly failsafe, the servers should be at different sites, requiring a decent sized broadband pipe, another expense

At a couple of client, I purchase at least two servers which are exactly the same, including the raid adapters. One of the servers is setup as the primary server for the company, the other is setup with non critical functions and programs. Should the primary server die, I can remove the disks from the secondary server and place the arrays from the primary server into the secondary, takes about 15 minutes; once the primary server is backup and running, the disks can be swapped in  reverse. At one of these sites, I have a removable disk carrier which is used for taking cloned disks offsite, mobile disks are swapped out every week. Might add the server room has a concrete floor, and 3 hour fire rate walls/ceiling, commercial surge protector, Halon equivalent fire extinguisher.

As to raid 1, You are much better with raid 10 for performance, raid 1 only reads from one disk of a raid 1 pair, and there is some overhead for writes to the second disk, as the writes must complete on the first drive written to, before the second drive gets the writes; on a non stressed server this has negligible affect as the cache absorbs this, but once under load the delays show up. In raid 10, you get a considerable increase in read speed, as it can read/write multiple disks at the same time. For performance stick with SAS drives.

"for higher safety I could make RAID 1 pairs between two MD1000's"

With raid 1, this is OK, with raid 10 or any other multi disk array it will lower your safety greatly. Should one of the MDs fail, you have just lost the array, with anything but raid 1 (in most cases). 

Since your new to this... mark/document your cables at all termination points, number the drives as to channel and slot number. Document your array configuration, so you can duplicate the configuration EXACTLY....there few things in IT as scary as not being sure of which drive go to which slot/channel on a production server at 3 am.    

Thank you very much for your kind advice.

I also agree that clustering two servers that depend on a unique data storage solution is a bad idea; for this I am considering (through testing) what seems to be a very intelligent and complete clustering solution:  DRBD (http://www.drbd.org/). Clustering 'twinned' servers (in hardware and storage space configuration), should anything at any point in the data transfer of one node fail (server, perc6 or RAID array), the other node takes over immediately.

RAID 10 will be my second choice if the above doesn't suit my goal, and yes, I see your reasoning there about the RAID 1 level disposition.

I would be very interested in hearing anyone's experiences with drbd.

Again, thanks. 

check out DoubleTake.....

http://www.doubletake.com/

One thing to be very careful of...get an eval version before you buy, if they do not offer a fully working  eval, do not buy it. Many of the server duplication programs use a great deal of server resources, especially the CPU, I tried 4 programs and most crippled the server performance, Doubletake was the only one at the time which did not tax the CPU to death.

Be careful to read any limitations, especially if you use Windows Active Directory