Link to home
Start Free TrialLog in
Avatar of Dan
DanFlag for United States of America

asked on

Is RAID 5 ok with ssd drives?

Hi, I was going to build server with 3 2 Terrible drives and put them in RAID 5 to get about 4 T of storage and fault tolerance.  

I remember allot of techs here didn't like RAID 5 due to the amout of time it takes to rebuild and that another drive can fail in that time.  Is that still true, or was it ever with SSD's?

The controller I'm using is a PERC H730P

Thanks a ton guys
Avatar of David Johnson, CD
David Johnson, CD
Flag of Canada image

always remember that RAID is no substitute for backups.
Why do you want to use raid? redundancy?  more iops?
The computing community is going away from raid i.e.
Storage Spaces Direct, Clustered Shares, VMWARE's storage solution, and so forth.
RAID 5's limitation is that error correction on recent drives isn't good enough when one creates raid arrays > 12 TB. You are almost guaranteed to have an uncorrectable error and have some corruption when rebuilding.
With less than 10TB arrays then RAID 5 is ok.
Do remember that usually when building a system all of the drives have the same powered on hours. So if a drive fails after 10K hours then the remaining drives are at the same lifetime and going to be stressed when rebuilding. This increases your probability of having an additional drive fail.

SSD's when they fail they usually fail catastrophically the drive is either good or bad and not failing. One can use something like HDDRegenerator/Spinrite on a spinning rust and recover enough to get you up and going.. SSD's not so much
ASKER CERTIFIED SOLUTION
Avatar of Tom Cieslik
Tom Cieslik
Flag of United States of America image

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
Avatar of Dan

ASKER

Thanks David, I know it's not a backup.  The client wants speed and redundancy, so if one failes, he is still OK, I suggested ssd's for the speed.  He only needs 4 T of data, I looked at mirroring, but 4 T ssd's are stupid expensive, so I figured three 2 Ts would be OK with RAID 5.

So it sounds like that is fine since it's under the 10 TB of space, is that right?

Thanks again.
Get two 2TB SSDs and either do RAID 0 for even faster speeds, or join them.  RAID 5 is only good for spinning disks under 1TB.  Anything over and you'll likely have a 2nd disk failure before a rebuild can complete.  With SSDs, RAID 5 can probably rebuild in an acceptable rate, but those SSDs don't last so long in server environments unless you use SLC nand instead of MLC or TLC.
Avatar of Dan

ASKER

Tom so with 4 2TB drives I get 4 TB of storage?  That would work.  Do you know if the PERC H730P does that?
Avatar of Dr. Klahn
Dr. Klahn

Side note:  No matter what controller and drives are in use, buy a spare controller and get it to the same firmware level as the one in use.  Then when the controller decides to stab you in the back you can bring up the spare in 10 minutes rather than wait 3 days for the replacement.

Same goes for the drives.  Buy at least one extra identical drive.  Remember that SSDs have a shorter cell life these days and when one drive goes (due to wear exhaustion), they are all probably going to go in rapid succession.  Alternatively, over-provision the drives, e.g. use 1 TB on 2 TB drives.  It is claimed that new drives won't fail due to wear exhaustion without clear warnings but this is a prediction that won't be known true or false until some years have passed.

And as David says, RAID is not a substitute for backup.  If there is an amazing power surge, if a rat gets into the cabinet, if an unhappy employee takes a hammer to the server, or (much more likely) ransomware gets into the system and encrypts all data files, then clean, daily, air-gapped backup is required.
I am going to disagree. There is nothing wrong with parity RAID on SSD. I would still use a RAID controller with on-board cache. The URE issues with SSD are orders of magnitude less than HDD, so the reasons why parity RAID with large HDD sucks don't apply. URE are lower, and random reads and writes are so much faster that your rebuilds are fast. I wouldn't hesitate to put 3 enterprise SSD on a PERC H730P in RAID 5.

While yes there are FT data protection schemes other than RAID, that doesn't mean that RAID is obsolete.

I have yet to have an enterprise SSD fail on me. SLC drives are extremely rare these days. It's MLC and going to TLC, and even QLC is now available. The early drives had write endurance problems. That is now only a problem for some niche use cases.
I have yet to have an enterprise SSD fail on me. SLC drives are extremely rare these days. It's MLC and going to TLC, and even QLC is now available. The early drives had write endurance problems.

I must respectfully disagree.  Unless you are referring to the very first SLC drives without wear leveling firmware, SLC flash has endurance of 100,000 write cycles.  MLC has endurance of 10,000 write cycles ... a reduction in usability by a factor of 10.  TLC has endurance of only 5,000 write cycles (the writer below is unable to call it anything other than "OK") and QLC will be half of that, which might reasonably be called "Fair."  Reducing endurance by a factor of 40 from 100,000 ("Excellent" down to 2,500 ("Fair") is IMO not improvement.

What makes TLC and QLC usable is over-provisioning, and even that does no good when the error correction is unable to correct a cell error.  That does happen and it happens with increasing frequency as multi-level cell voltage levels become more and more poorly defined over time.  And unlike a hard drive where sectors go bad one by one, when an SSD begins to fail the whole thing fails at once.

I might consider 2-level eMLC for an enterprise drive but garden variety MLC, TLC and QLC IMO should be considered consumer grade.

https://www.enterprisestorageforum.com/storage-hardware/slc-vs-mlc-vs-tlc-nand-flash.html
It's probably cheaper to buy a single 4TB ssd and a cheap 4TB or 5TB HD, maybe external, to back data to and keep offline until needed.  The extra "encumbrance' of RAID, plus multiple SSDs means a slight overall increased failure rate that's compensated by the redundancy.  You'll use less electricity in the long term and have less complexity.
Avatar of Dan

ASKER

Thanks all, I'm going to do as Tom said and just bite the bullet and get two 4TB drives, I don't think RAID 5 is worth it.  I leared a ton here, thanks all.
Hi Dan
Any PERC H730P can create Raid 10 as long as you'll have backplain available for hot swap-able disks or available connectors on controller or MB itself.

If you going to use 4 disk 2TB each you'll get Raid 10 with 4TB storage or you can use 2 disk 4TB each to get Raid 1 Mirror. That's enough for security
Avatar of Dan

ASKER

cool, thanks.
"I wouldn't hesitate to put 3 enterprise SSD on a PERC H730P in RAID 5"

One thing to note here is that the drive you put on this controller matters. You may not be able to just throw in a some Samsung EVO's and expect miraculous performance and reliability. Some "consumer" drives work and play nicely with the enterprise Dell controller, some don't.
I would add that after doing some research on my own controllers, that the H330 does NOT have an onboard XOR hardware engine, so it really shouldn't be used for any parity RAID.
No, the H330 is probably only good for pass-through of SAS drives to the OS.
Hence, you really should get a single 4TB SSD or use two 2TB SSDs and concatenate them.  Then use an external 4TB or 5TB HD for offline backup.  Make sure you back it up and take it offline. (disconnect the drive and unplug the power.