SSD RAIDs Revisited

[LunarMist]

Tyically I implement a SSD boot drive and a second SSD for certain purposes, such as favorites, Outlook, PTGUI temp files, etc. The rest of my SSDs are in laptops or PSDs.
However, with the modestly-priced 1TB SSDs available today, I'm exploring the feasibility of using SSDs in the desktop for data storage. I can live with a non-redundant RAID for 2-4 SSD drives, but I recall various problems in the earlier days of RAID. Extra performance is not necessary, but the array must at least as fast as a single drive. Are there any compatibility issues with modern SSDs in RAID 0 or is JDOB more desirable? Thanks.

 

[Mercutio]

The absolute ONLY SSD RAID setup that has been qualified for any configuration at all is RAID0 on Intel 7-series boards under Windows 7 and 8. That's it. As far as I know, nothing else passes TRIM commands, which means anything you'd choose to do would result in swift memory cell death.

 

[ddrueding]

I completely believe what Merc has posted above, but I've run Vertex drives in RAID10 off 3Ware controllers without performance penalty; in fact it was pretty darn quick. Haven't done anything like that in a long time, the drives are big enough now that I don't bother. What sort of total capacity are you looking for? I also don't know if softRAID in the OS allows passing TRIM commands.

 

[LunarMist]

I completely believe what Merc has posted above, but I've run Vertex drives in RAID10 off 3Ware controllers without performance penalty; in fact it was pretty darn quick. Haven't done anything like that in a long time, the drives are big enough now that I don't bother. What sort of total capacity are you looking for? I also don't know if softRAID in the OS allows passing TRIM commands.

I have 2x1TB for laptops/PSDs and will probaby get another one, so 3TB.

 

[LunarMist]

The absolute ONLY SSD RAID setup that has been qualified for any configuration at all is RAID0 on Intel 7-series boards under Windows 7 and 8. That's it. As far as I know, nothing else passes TRIM commands, which means anything you'd choose to do would result in swift memory cell death.

What do you mean cell death? The PSDs don't have TRIM either and I don't know about USB 3.0. I thought the modern drives cleaned themselves up autonomically. Is there some way to manually clean up the drives that does not involve some crappy boot disc and a native port?

 

[Bozo]

Windows 7 issues a TRIM command after deleting files. Intel updated their software to issue the TRIM command to RAID 0 setups, but only on their control chips and and only on certain Intel SSDs.

 

[LunarMist]

I undersatnd, but did not think it was so critical nowadays. I'd be breaking the array a few times a year when traveling anyway. Unfortunately Intel has no appropriate 1TB drives, so I had to go with the Crucial brand. They don't seem to have any optimization software as Intel or Samsung do.

 

[Mercutio]

I completely believe what Merc has posted above, but I've run Vertex drives in RAID10 off 3Ware controllers without performance penalty

In my experience, SSDs quickly in soft- and hardware RAID configurations, though performance while they're working is excellent. Yes, even RAID1. I've experimented with building small servers with SSD RAID1 setups to maximize performance and found that I was having to break a lot of arrays and RMA a lot of drives.

 

[LunarMist]

OK, I give up.

 

[time]

No idea what absence of Trim has to do with drives dropping off your array etc.

It's not that long ago that most SSDs couldn't take advantage of Trim anyway. Yes, it can reduce fragmentation, but so can the drive's own housekeeping as long as you leave it plenty of free space.

 

[Chewy509]

No idea what absence of Trim has to do with drives dropping off your array etc.

That was what I was thinking, TRIM support should have nothing to do with it.

My guess is that the Garbage Collector on the drive is kicking in and causing higher than normal IO latencies to occur, that the RAID controller is interpreting as a dying/misbehaving drive, which the RAID controller then drops the drive... (Same reason why a lot of Green Drives don't play well in RAID arrays, the time to spin up is causing the controller to think it's got a bad drive).

Also note, we have a few SSDs at work that exhibit weird pausing issues (as in up to 30 secs pause) every now and then, and we can't find anything wrong with the drive, and the manufacturer just claims it's the GC kicking it...

 

[CougTek]

Also note, we have a few SSDs at work that exhibit weird pausing issues (as in up to 30 secs pause) every now and then, and we can't find anything wrong with the drive, and the manufacturer just claims it's the GC kicking it...

It's a known issue with SSD using a Marvell controller. Many Crucial and Plextor drives exhibit this behavior.

 

[LunarMist]

Is it acceptable to use a dynamic disk on three SSDs and Spandex mode to create a 3TB parturition? RAID 0 is slower than a single drive.

 

[CougTek]

Is it acceptable to use a dynamic disk on three SSDs and Spandex mode to create a 3TB parturition? RAID 0 is slower than a single drive.

You won't go to jail if you do it, but it looks like a nice way to increase your odds to lose data.

 

[LunarMist]

Why, is it an SSD or Windows problem?

 

[P5-133XL]

Windows! The file system does not like it if a big chunk disappears. So, just like Raid-o, if a drive fails in spandex mode then you are likely to lose most if not all data.

 

[LunarMist]

Windows! The file system does not like it if a big chunk disappears. So, just like Raid-o, if a drive fails in spandex mode then you are likely to lose most if not all data.

Oops, I mean spanning. Damned Androic checker...
So it's only 3x less reliable? Aren't SSDs more reliable than HDDs, though?

 

[Chewy509]

Aren't SSDs more reliable than HDDs, though?

In which context?
Overall, in theory yes... but in realty, I suspect the failure rates are about the same.

 

[ddrueding]

To date I've had one SSD fail that wasn't due to some external factor (blown PSU, crazy thermal conditions, etc). This is way better than the HDD failure rate I used to deal with.

 

[P5-133XL]

Oops, I mean spanning. Damned Androic checker...
So it's only 3x less reliable? Aren't SSDs more reliable than HDDs, though?

You asked it if was acceptable, not if more reliable than raid-0. I like SSD's and for many uses I will agree that they are likely more reliable.

That being said, spandex mode creates a multiple drive dependent situation where if any one of multiple SSD's fail you will lose data and that is unacceptable to me. For a server situation it is multiple redundancy that is the minimum acceptability; For a consumer, single points of failure are the norm; Never, is multiple points of failure a route to acceptability unless the data means little to none.

 

[LunarMist]

I decided to use only two of the SSDs, rather than three. I wonder how much the performance will deteriorate when the first drive is full and data spans to the second drive.

 

[P5-133XL]

Personally, I do not see where you get much additional speed from using spaning mode.

My preference for making a huge drive is to organize the file system so it splits the data reasonably and the mount multiple drive(s) into a folder(s) rather than a drive letter. I recognize that there are limitations to that method because sometimes it is outright hard to impossible to organize the data to fit efficiently.

A second alternative, depending upon the OS that you are using, is to have separate drives but use libraries to organize the data into a appropriate groups.

 

[ddrueding]

If you are going to span you may as well stripe

 

[LunarMist]

If you are going to span you may as well stripe

I tried striping three different ways (including the LSI hardware RAID 0), but real-world throughput is lower than a single drive. It looks fine on the stupid benchmarks though.

 

[LunarMist]

Personally, I do not see where you get much additional speed from using spaning mode.

My preference for making a huge drive is to organize the file system so it splits the data reasonably and the mount multiple drive(s) into a folder(s) rather than a drive letter. I recognize that there are limitations to that method because sometimes it is outright hard to impossible to organize the data to fit efficiently.

A second alternative, depending upon the OS that you are using, is to have separate drives but use libraries to organize the data into a appropriate groups.

The reason I want the two SSDs in an array is to keep the partition at a usable size (960GB is too small for anything) and to avoid any need for additional letters. I also need to back up those drives. In total there are 52TB of drives on this system, so it is a hassle to make a significant change. Basically I rep

 

[CougTek]

Damn! Looks like Mooner had a heart attack before finishing his post.

 

[LunarMist]

I'm not sure what occurred.