Upgrading server from HD to SSD drives

Custom built pc / Ga-z97x-ud3h
June 20, 2019 at 07:17:41
Specs: Windows Server 2012, 8GB will be 16GB
The server uses 3 drives in Raid 5, each drive is 1TB but the total usage is only about 250GB after 5 years. I am for now assuming that the drives are SataIII drives. My question is If you replace one drive at a time with a 1TB SSD Sata drive, will it rebuild the array that way and if you replace the next drive the next day until all drives are replaced, will this be a nearly zero down time (shutting down once a day in morning or evening for enough time to physically exchange drives)? Is this possible? Is this practical? I was told that the system should be backed up, the array should be remade with two 1TB drives in Raid2 (mirror) and rebuilt from the back up. It was said that this would take 6 hours of downtime and labor. This seems excessive. A third drive would be considerably cheaper than the excessive labor cost and the downtime prohibitive. If I am right, we could make the drive swap ourselves. Please explain completely.

You have to be a little bit crazy to keep you from going insane.


See More: Upgrading server from HD to SSD drives

Reply ↓  Report •

#1
June 20, 2019 at 08:33:22
I've upgraded the hard drives on IBM servers this way many times, so I don't see why it should be any different upgrading hard drives to SSDs, as long as their capacity is equal to, or greater than, the hard drives. This was more than 10 years ago and I would assume that modern RAID controllers are as capable, or more so. The only reason I can see to adopt the procedure that you have been recommended is if you positively wanted to switch from RAID 5 to RAID 2. (Are you sure that you mean RAID 2 rather than RAID 1 - I thought that was outdated?)

Having said that, I practiced the operation on an identical spare server first, just to be sure (I had very understanding employers who appreciated that the expense of a spare server, that was hardly ever used, was a good investment). And you should, in any case, have full backups of the system.

Do you have a spare server, or even just a spare computer with the appropriate capacity? If so, the safest plan would be to copy all the data onto that computer, switch it with the server, at your leisure upgrade the server, copy any changed data from the temporary machine to the server and switch back again. Whatever OS you are running there should be utilities that allow you to copy only changed files, which makes the changeover very quick (obviously both machines have to be offline whilst you copy the final changes before switching.) It's always more reassuring to work on a non-live machine avoiding the stress involved otherwise.


Reply ↓  Report •

#2
June 20, 2019 at 10:40:18
Fingers: I am for now assuming that the drives are SataIII
Is the RAID backed by a hardware controller? If so, that might be a bad assumption. Verify drives aren't some flavor of SCSI before proceeding.

Fingers: If you replace one drive at a time with a 1TB SSD Sata drive, will it rebuild the array
In theory. It'd be a bad RAID controller if it didn't, and you should probably seek a different controller. Consult your controller's documentation before proceeding. I'd like to see the controller support TRIM, but the industry in general doesn't seem to think it's required. So long as the drive's GC can keep up with the application's writes, you shouldn't have an issue there.

Fingers: Is this possible? Is this practical?
The RAID controller should rebuild the array in the background, and the only thing the rest of the system should see is some performance degradation while the rebuild happens. As far as what the application sees, nothing on the RAID presented disk actually changes. If the array supports hot swapping, I'd probably wait for a slow period and do the swap then to get zero downtime.

Fingers: the system should be backed up
Always good advice in any situation.

Fingers: Raid2
RAID 2 didn't make it out of initial testing of the spec. Question their insistence on RAID 2, and question their technical advice if they don't provide a satisfactory answer. Refresh yourself on RAID 1+0 / RAID 0+1. Question their technical advice if their definition of "RAID 2" is actually RAID 1+0 / 0+1.

Fingers: This seems excessive.
Are they trying to sell their services to do the swap? Might explain the 6 hours quote. "No, no, no! Don't do it yourselves! It has to be done by qualified professionals! Like us! We'll get you back up and running at the low, low cost of $$$ / hr."

How To Ask Questions The Smart Way


Reply ↓  Report •

#3
June 20, 2019 at 11:13:28
Just to emphasize once more - make sure the replacement drives have a capacity at least that of the originals. Don't just rely on "1 TB", check the exact capacity.

Reply ↓  Report •

Related Solutions

#4
June 20, 2019 at 13:06:49
Sorry, I was thinking wrong when I said RAID 2, I assumed that raid 5 being 3 drives, going to 2 drives (mirrored) but got confused. Now I remember raid 0 is the no-no (no protection) and raid 1 is mirrored so I guess it would be Raid 1.
They were quoting a more expensive hot swap drive (which I do not think it is) from Axiom at over $500. each but Samsung and Intel Enterprise SATA drives are in the range from $175. to $210. each. My thought is if they need to rebuild the array from scratch because going from an array of 3 to 2 due to the price of the drives, With the less expensive drives staying with three is more than reasonable, especially if the downtime and professional labor charge would be less.
What are your thoughts on these drives:
https://www.newegg.com/p/0AY-00PN-0...
vs:
https://www.newegg.com/samsung-860d...
or:
https://www.newegg.com/intel-dc-p36...
or similar?

You have to be a little bit crazy to keep you from going insane.


Reply ↓  Report •

#5
June 20, 2019 at 15:37:14
That's an entirely different can of beans altogether. You're not talking about rebuilding an array, you're talking about tearing down the existing array, and building a new RAID array with the old data. To do that, you'll need to get both arrays up at the same time and do an OS-level disk copy from the old to the new array. The copy will probably take a few hours. After that, you can remove the old RAID 5 array, and configure the new RAID 1 as the replacement disk. Other option would be to backup the RAID 5 array, destroy it, build the new RAID 1 array, and restore RAID 5's data to it. In either case, you'll have to deal with the resulting complications, especially if Windows also lives on the array. Expect multiple hours of downtime for the data copy and cleanup. Is the downtime and man hours cheaper than getting a third drive and keeping the RAID 5? Don't know; you'll have to do the math yourself.

As for the drives, first link doesn't work, second link's specs looks like a consumer grade drive, third link uses U.2 and not SATA.

How To Ask Questions The Smart Way


Reply ↓  Report •

#6
June 20, 2019 at 16:50:22
Sorry, I will try again on the links

https://www.newegg.com/p/2U3-001B-0...
https://www.newegg.com/samsung-860d...
vs
https://www.newegg.com/p/0AY-00PN-0...

The raid one option was their suggestion with the complete rebuild but the down time seems excessive and the labor at 6 hours (most downtime) is not a workable solution. It was based on the pricier drives.
With a more reasonably priced drive I felt that rebuilding it would leave little or no downtime and less labor time. They are putting forth that the swap and rebuild from conventional to SSD a drive at a time would cause all kinds of problems. If the drives need to be at least the same size or larger then that is not a problem with the 1.6TB drives being half again larger and in the same price range. It is an HP server running Windows Server 2012, do you know of anything that would create the kind of issues they are telling me about if we stay with 3 drives in raid 5 and use one of these SSD drives because I assume that the labor here would be higher than the third drive if it drastically cut down on the time needed rebuild the array from scratch. Also the downtime was supposed to be 6 hours their way and I am assuming that the drives can be put in one at a time and left to rebuild and repeated another time until all are replaced. This should keep downtime to less than a half hour maybe three days in a row if I am right. This is what I was expecting it to be but they are warning against it. What do you think?

You have to be a little bit crazy to keep you from going insane.


Reply ↓  Report •

#7
June 21, 2019 at 09:30:37
Yeah, one of your links is apparently tripping the spam link filter. Also, you linked to a drive that wasn't quite 1TB. Both of the drive's specs seem like consumer grade drives, so you'll have to decide if the possible I/O hit is worth the difference in price.

Data copying is expensive, especially when we're at the GB to TB scale. As for their concerns, I'd have to know what they were before I could respond. If their concerns are that it would, "cause problems," I'd have to respond with, "Such as?" If they feel it would be a bad idea but don't have anything definite, they should say so. That way, I could take it under advisement, and make sure I have a back out plan for the RAID 5 rebuild. Specifically, if the rebuild doesn't start, I could reinstall the old drive (making the array healthy again), and reconsider my options. Basically, don't wipe the old RAID 5 disks until the array is rebuilt with SSDs.

How To Ask Questions The Smart Way

message edited by Razor2.3


Reply ↓  Report •

#8
June 26, 2019 at 21:09:48
A better solution was recently suggested with little downtime.
Leave the existing RAID array to run the operating system and build a new mirrored array and just move the data files from the old array of hard drives to the new array of SSD's. Along with additional memory this should solve the speed issues with little downtime and even less risks.
I do not know yet if the current raid controller will support the second array as it is or if it is the best option if it does but I have asked for a formal proposal so I can help management decide how best to proceed.

You have to be a little bit crazy to keep you from going insane.

message edited by Fingers


Reply ↓  Report •

Ask Question