Convert RAID from Mechanical Disks to SSD?

Question

Any issues with converting a Mechanical Disk RAID to SSD by replacing the drives one-by-one?

I couldn't find anyone online who has done this, so let's make this the place to post about it.

If you have successfully converted a RAID from Mechanical Disk to SSD, please come back and post your specs. (serverfault doesn't ask for registration, so it's really quick to post.)

FYI, I'm running an LSI MegaRAID 9260-8i with 8 3TB ST3000DM001 drives in RAID 6, so I'm a few years away from affordable 3TB+ SSDs. I've had 2 drives marked as Failed over the past 3 years and each time it took 8hrs to rebuild a drive. Of course backup first, and I'm aware of the risks to the other drives as I replace and rebuild each with SSD.

At the time of this post, 1TB SSDs can be bought for ~$350 each so people with smaller/less drives in their arrays may already be thinking about this. Does a 1TB SSD have the same capacity as a 1TB mechanical drive, or would it need to be replaced with something larger, like a 1.2TB SSD which is also available? Since oversizing the SSDs is an expensive proposition, what is the best way to confirm the actual max capacity of the drives? (Best practice is to use 3-5% less than full capacity to allow using a different model of the same size if yours is discontinued, but I know a lot of people don't do this.)

Solution 1

I have done something like this a few months ago.

A Supermicro-Server with Intel ICH8R Raid-Controller had 4 Disks in 2 Logical Volumes

• Volume 0: 2xWD Re 250GB SATA - Raid 1
• Volume 1: 2xWD Re 500GB SATA - Raid 1

As Volume 0 was the System-Drive for a SBS 2008 and was suffering IO-Performance we decided to replace The drives for Volume 0 with SSD's.

We did this the brutal way. Switched off the Server, Replaced the first drive (WD Re 250GB) with a Samsung SSD 850 PRO with 512 GB. After turning on the Server the Raid controller was instructed to rebuild the array on the newly added Drive. After the rebuild finished we repeated the procedure with the second 250 GB Drive.

No other changes were made. The Performance of random reads (Database) increased appreciable.

You may ask why we choosed 512GB SSD's as replacement for 250GB HDD's? That was just because of overprovisioning and extend the life of the SSD.

Solution 2

We just did the same thing Martin Seitl mentioned on a development server, with an LSI SAS1068E, from two Hitachi Ultrastar A7K1000 SATA disks to two SATA Samsung EVO 840 drives (and we installed the second firmware fix for the EVO 840 performance degradation...). There are, however, a few things to note:

• Be sure your controller can mix SATA and SAS, if applicable. Our 3Ware 9650 8i controllers can't. LSI appropriated 3Ware, and from the type nr, I would guess that your controller suffers from the same issue.
• Are your current partitions 4k aligned? I'm not sure if it's necessary for SSDs, but I suspect so.
• RAID controllers don't support trim. You're not able to do anything about it, but at least be aware of it. You may get write-amplification over time. Over-provisioning like Martin did is indeed smart, but not perfect.

As for disk sizes; I've noticed that from about 500GB, disks from different manufacturers are sized exactly to the byte (but I don't know if this is a real standard). This can be easily verified in the specs of the drive, though. But bear the over-provisioning in mind.

Comments

[JimNim]

+1 - LSI cards certainly don't support this. You would only see this kind of capability in a RAID enclosure, and even then there would be other methods recommended to accomplish your end goal.

[Halfgaar]

@ewwhite never mind, my mistake. I re-check what the drives were.