How Sensitive Are R.A.I.D. Arrays To Power Cut-Offs?

[wpcoe]

I want to put two 60GB SSDs in a RAID-0 array. The price is about the same as a single 120GB SSD, but the speed would be quite a bit faster. I'm familiar with the THEORY of RAIDs, but never operated one.

I asked a Canadian friend of mine who is a network admin and deals with RAIDs at work (and has one on his home computer) if there were any hassles or negative issues with running a RAID, and his one-line (typical) reply was: "No problems. .RAID is very reliable.. they never break or corrupt, unless you incorrect shutdown your computer."

That set off alarm bells for me, thinking of the frequent power outages and dips here, where my computer shuts down. Yeah, I could (and should!) get a UPS, but unless I get one with auto-Windows-shutdown (i.e. expensive) it wouldn't matter when I left my rig running unattended -- as I frequently do when I have large downloads in progress -- a power outage would still abruptly "power off" my computer when the UPS battery died.

So, how practical is it to run an unattended RAID-0 in Thailand? I'm in Hua Hin, and it seems I have more power "snaps" here than I did when I lived in Jomtien. Just this week, I think I have set the clock on the microwave 5 times.

Are RAID arrays prone to break down if the computer is not shut-down properly? If so, would it probably involve data corruption, or is it just a matter of re-initializing the RAID on next boot up?

[bobl]

Some RAID controllers have battery backup to help them in the event of a power failure. Most don't (even the ones that do, it's an add-on "option").

Long story short - there is a significant likelihood of your data becoming unusable on a raid, in the event of a power failure (i.e., higher likelihood than with a non-raid setup)

Difficult to quantify though. Another way to put it would be to compare it to a standard PC. 99 times out of 100 your pc might survive a power cut and/or at least be "fixable" with a chkdsk. A PC with a raid is very unlikely to survive 99/100, nor would it be nearly so easy to recover the data in the event of a disaster.

[dave_boo]

I am going to have to respectfully disagree with bobl.

All recording media will suffer an error if writing when the power goes out. Remember the good old fashioned 5,25 or 3,5 floppies where they always said 'Don't pull the disk until the write indicator shuts off'? Same idea.

HOWEVER, your choice of RAID 0¹ does leave you open to a greater chance of hard drive failure. I think the statistic is 3% of hard drives fail in the first year; this is for the spinning platter version. If you have 2 hard drives, that means that your chance of failure jumps to 5%. Not such a big deal if you keep backups (which I assume that with 120GB of space only you are going to), but if either drive fails than the whole array is hosed. RAID 0 works by striping the data across the drives, so with a two drive array if either were to go out than you'd be missing approximately 1/2 of each and every file.

If you buy a UPS, most, if not all, have a super annoying alarm that goes off when they loose power in. And as you said, you can configure Windows to shutdown if the UPS is a good version (has either an RS-232 or a USB connection).

1) I hate that people call striping "RAID 0". RAID is a Redundant Array of Inexpensive Disks. While striping is used in a RAID array (every level from 2-6) if there is no parity either through mirroring or parity (RAID 2, 3, and 4 use dedicated parity disks and RAID 5 and 6 stripe the parity across all disks with RAID 6 creating two parity blocks for each block of data written).

Edited February 9, 2012 by dave_boo

[Crossy]

Will striping of SSDs give a noticeable performance improvement anyway? These things are blisteringly quick to start with.

[dave_boo]

Will striping of SSDs give a noticeable performance improvement anyway? These things are blisteringly quick to start with.

Of course it will!

But there is a bit of a question that I have; will he be able to use the TRIM support baked into the drives if they are in an array? There's only specific combinations where that is possible and if he is not using one of those performance will degrade over time. Also as an aside, although I do not think it's such a big deal with only 64 GB drives, if the OP is getting 6 Gb/s drives make sure that they are on 6 Gb/s ports to get the maximum speed.

[wpcoe]

But there is a bit of a question that I have; will he be able to use the TRIM support baked into the drives if they are in an array? There's only specific combinations where that is possible and if he is not using one of those performance will degrade over time. Also as an aside, although I do not think it's such a big deal with only 64 GB drives, if the OP is getting 6 Gb/s drives make sure that they are on 6 Gb/s ports to get the maximum speed.

It has been reported that TRIM will be supported within RAID in the next version of SRT. If the SSDs have competent garbage collection (which most do these days), TRIM is not so vital, anyway.

I would run the striped array off of native (Intel chipset) SATA-3 ports. My motherboard has supplementary Marvell SATA-3 ports that are documented to be hamstrung in their implementation, so I won't be using those.

I know that a power outage can corrupt any disk being written to at the precise time of the outage. With the striped array is there MORE of a risk that data will become scrambled (other than doubling the odds for two drives)? My friend made specific reference to shutting down the array properly. Is there something that's essential in the shutdown process -- which would be missing in a power outage -- that has a greater likelihood of rendering the array as unusable at next boot-up?

[dave_boo]

It has been reported that TRIM will be supported within RAID in the next version of SRT. If the SSDs have competent garbage collection (which most do these days), TRIM is not so vital, anyway.

I would run the striped array off of native (Intel chipset) SATA-3 ports. My motherboard has supplementary Marvell SATA-3 ports that are documented to be hamstrung in their implementation, so I won't be using those.

I know that a power outage can corrupt any disk being written to at the precise time of the outage. With the striped array is there MORE of a risk that data will become scrambled (other than doubling the odds for two drives)? My friend made specific reference to shutting down the array properly. Is there something that's essential in the shutdown process -- which would be missing in a power outage -- that has a greater likelihood of rendering the array as unusable at next boot-up?

I apologise; you were not specific as to the drives you were going to get nor the controller.

As you understand a powerdown during a write is going to cause a problem, I'm going to throw two more terms at you: write through and write back. Write through is the data being written to both the cache (RAM) and the destination (array). This takes more time but is 'safer'. Write back merely writes to the cache and when that is full or the OS tells it to does it get written to disk.

I am unsure how Windows handles this (granted here is a link, but I don't use Windows seems simple enough though); using a 'real' controller, as mentioned earlier with a battery backup, allows one to get the best of both worlds; the speed up of write back with the safety of write through assuming the power outage does not last longer than your BBU.

[Digitalbanana]

>Just this week, I think I have set the clock on the microwave 5 times.

Don't think I've seen more than that in past two or three years. I'd get a UPS,

[wpcoe]

Don't think I've seen more than that in past two or three years. I'd get a UPS,

For my whole house? (just kidding)

My DVD player's not plugged in unless I'm using it, so that clock doesn't flash, at least. Now that I think of it, a "typical" UPS is probably good for 99% of the power disruptions. Usually it's just a one second (or less) outage, but that's enough to turn everything off in the house. Of course, most things pop right back on, but my computer and microwave clock are the two that don't.

In the ten months I've lived in Hua Hin, I think there's been "only" two long-term outages where a UPS battery would run down. Seems like over in Jomtien there would be maybe one per year, not counting the couple of planned (and announced) outages when PEA was restringing power lines, or when the condo planned some maintenance.

[wpcoe]

As you understand a powerdown during a write is going to cause a problem, I'm going to throw two more terms at you: write through and write back. Write through is the data being written to both the cache (RAM) and the destination (array). This takes more time but is 'safer'. Write back merely writes to the cache and when that is full or the OS tells it to does it get written to disk.

I am unsure how Windows handles this (granted here is a link, but I don't use Windows seems simple enough though); using a 'real' controller, as mentioned earlier with a battery backup, allows one to get the best of both worlds; the speed up of write back with the safety of write through assuming the power outage does not last longer than your BBU.

Thanks! I'm going to take some time to read about that. My mind is starting to fog over right now.