RAID-1 SSD protection schemes

L

lightp2

Gawd
Joined
Oct 28, 2009
Messages
672
RAID-1 SSD protection schemes

This post targets consumer environment. Enterprise Storage has advanced technology to deal with such issues, I think.

A. This is continuation of idea on different thread few days ago

1. In RAID-1, if you wrote X to Disk-1, you also need to write X to Disk-2
2. So if you wrote 20GB data to Disk-1, you probably need to write 20GB data to Disk-2
3. So if you wrote 40TB data to SSD-1, you probably need to write 40TB data to SSD-2
4. So if SSD-1 failed due to write endurance failure, SSD-2 will probably fail soon after the 1st one if both SSD are same model/capacity/partition-setup, in standard simple RAID-1 mirror setup.

B. Counter-measures

1. 1st is to use different spec SSD. Example SSD-1 is SLC. SSD-2 is MLC.
2. 2nd is to use different capcacity SSD. Example SSD-1 is 160GB SSD-2 is 80GB
3. 3rd is mix of different brands/models. Example Intel X25-M G2 160GB and SandForce-based 120GB

The basic idea is by having different specification leading to uneven pairing, it will modify the dynamic of write failure sucession in SSD RAID-1 scenario. This will avoid your RAID-1 to fail completely in short amount of time after encountering write endurance failure.

There is another way to address this issue.

1. Based on idea from Linux MD RAID. I have read-up but I have no experience with actual setup for now.
2. In Linux MD RAID, in RAID-1, you can have special setup.
3. Example, 2 Disks mirror, designate one disk as write-mostly/write-behind.
4. In that case, MD RAID will treat it as such (write-mostly), it will not read from it unless no choice.

5. So RAID-1 mirror
5a Disk-1 = SSD
5b. Disk-2 = Mechanical SATA disk (write-mostly)

In this case, Read IOPS comes from Disk-1 SSD so you enjoy full benefit of SSD IOPS
Disk-2 serves mostly as protection mechanism. In case SSD failed due to write endurance, you can still continue with the mechanical disk. This hybrid SSD/Mechanical RAID-1 setup may be useful in certain situations where you want improve performance but limited fund and you are also worry about failure issues mentioned in section A.
 
Why not just check every month or so how your SSDs are doing regarding write endurance? If you know it's at 90%, you can take precautions. So what are you afraid of, really?

If you really want RAID1 on SSDs then just pick similar ones which will fail at about the same time; then both can be replaced by newer models.

The stuff that goes on an SSD in consumer case would be stuff that does not need to be backupped; i.e. windows directory; installed programs (you have to reinstall anyway); pagefile; system restore; etc.

So the easiest would be to just move (or remap) your my documents folder to store on HDD instead. And keep the SSD as system disk; which may not have to store any important user-data on it.
 
If you are worried about that type of long-term failure... get SmartSync Pro or Second Copy and have it archive whatever you keep on your OS/SSD to a mechanical drive or another SSD. You could even have it do it monthly if you want and lower it by even more writes. I personally am more worried about the data on my mechanical drive, since the stuff on my OS/SSD I can re-install for DVDs in less than an hour or two. Now if I lost my documents, pictures, e-mails, etc.. that would suck! So, I also use SmartSync Pro to backup to my NAS which is RAID1 Mech drives.

Unless you have a 250gb+ SSD and keep everything on it there's no point to RAID1 it IMHO. Backup what you need off their at an interval your comfortable with and be done.
 
I would just backup the SSD. I do not see the point of wasting a second ssd on raid-1 it would be better served for extra ssd storage. Do your backups to a hard drive instead.


I personally am more worried about the data on my mechanical drive
Click to expand...

Same here, much higher chance of loosing data with a mechanical disk.
 
RAIDing SSD's for the most part is just silly.

Unless you need some crazy fast transfer speeds, its nearly pointless. The speed increase comes at the cost of latency then. Writing to the drives on a RAID 1 setup will take longer, reads, shorter - in theory. But since the SSD's will push you're average controller you would be hard pressed to find any read benefit.

Get a better SSD, and then backup every day (incremental) at some point, with a full backup once a week, then once its full delete the oldest full backup - onto a 500GB drive. This means you'll have at least a months backup PLUS you're latest incremental.

IMO thats the best "cheapest" solution. 500GB drives are real cheap now, and backing up a 128GB SSD (since you're not gonna fill it past 100GB) you get 4 weeks + incremental. Heck, get another 500GB shove it in an external case / dock and backup of to that for off-site storage if you need it.
 
I'm going to have to agree here. The cost/performance ratio of SSDs and the cheap cost of a mechanical hard drive doesn't make sense to RAID1 an SSD. You perform your nightly or at least weekly backups and run with it... WHS nightly backups take less than 5 minutes for me.
 
OK. I agree all of you are absolutely right for home usage.

I am not trying to be stubborn, I will modify my claim slightly.

1. Currently there are a lot of Linux servers out in the field and Internet

2. A lot of them are on Linux MD software RAID. (especially those entry-level with 2-SATA disk connections type)

3. You will find portion of them on 24x7 duty on Linux software RAID-1. You will also find many deployed to data-center and work remotely.

4. In future when SSD has reached reasonable stability and cost factor satisfactory, generic server deployment will become common. Issues discussed in my post may become consideration.

I am still trying to cook up other schemes for such scenario.
 
Last edited:
So example, some entry-level Linux servers are running Internet Web App and MySQL.

The database has much less Write but higher Read IOPS demand. Since these are usually slim 1U with only 2 disk running remotely, using the hybrid SSD/Mechanical RAID-1 maybe useful in certain situation.

Specific example, say database-based Web Product Catalogs. WebSite maintainer update product items maybe one or twice per day. The rest is aggressive searching by internet user (Read IOPS). Multiply this on webhost and I believe you can see some benefits. The RAID-1 ensure your webhost keeps running even if you encountered SSD write endurance failure, (perhaps) with server residing in datacenter. Some admins cannot say just let it down for 1 or 2 days.

There are concern of MLC for long-term reliability in such environments. I am not doing website so this is only thinking. Maybe other can comment.
 
You didn't mention anything to do with servers in the OP....

I would only use MLC with trim support in mission critical stuff.

Else I'd drop down on some 15 or 17k SAS drives or SLC enterprise stuff.

When its mission critical, you don't fuck about.
 
You must log in or register to reply here.
Back
Top