Bad luck with SSD's - am I the only one?
-
So… last evening the 16GB SSD drive in my pfSense box died. This is the third dead SSD in a year! I'm running pfSense 2.0.x full install (over the last year 2.0.1 -> 2.0.2 -> 2.0.3) with Squid, Dansguardian and some custom ipfw rules. Here's the deal with each SSD...
First one was a 16GB that I bought used on eBay. It was used - but a name brand. It started getting corruption (I think it hand multiple bad block messages) within a couple of months. Thought - hmmm... I'll buy a new drive next time. The second drive was a Kingspec and I believe it was 16GB also. Occassionally it would start having write errors like crazy. I'd reboot the box, the errors would get fixed on bootup and everything would work fine for a week or two. I just blamed it on the fact that it was a Kingspec and decided I should have bought a name brand drive. So the latest drive... It was a 16GB Kingston SSDNow and it had the weirdest failure yet. The little plastic tab around the pins for the power connector physically broke off!!! while the drive was in the box running!!! It baffles me - there was no pressure on it or anything. Something must have shorted when the connector broke and now the drive is completely dead.
So... at the moment I'm running on an old 80GB 3.5" drive. My original thinking was that an SSD should be much more reliable - no moving parts, right? Likewise, modern wear leveling algorithms should give years of service. But my experience so far with an SSD for pfSense has been terrible. On the other hand, I haven't had any problem with the one I put in a laptop. Is an SSD just a bad idea for pfSense?
-
Stop buying ancient drives. The first couple generations of SSDs, particularly those based on a JMicron controller, are absolute trash, especially under a write-heavy load.
I've not had a single Intel SSD fail and some of them have had hundreds of TBs of data written to them.
-
Stop buying ancient drives. The first couple generations of SSDs, particularly those based on a JMicron controller, are absolute trash, especially under a write-heavy load.
I've not had a single Intel SSD fail and some of them have had hundreds of TBs of data written to them.
If buying a quality, new SSD is the answer, I'm willing to do that…
On a related note - should I get an SSD that is larger than what I really need? For my home network the 16GB SSD has plenty of space - even with a decent Squid cache. However, should I get a 64GB so that the wear leveling algorithms have more space to work with?
-
I think we've heard more reports of Kingston SSDs driving than anything else by far. I wouldn't trust those very much at all.
After that, we've heard sporadic reports of OCZ and Crucial drives giving out here and there.
I've only had one person they that an Intel SSD failed them. They're the best by far from all measures I've seen.
SSDs sound great on paper but the tech needs more time to mature before I trust it.
-
I've only had one person they that an Intel SSD failed them. They're the best by far from all measures I've seen.
Samsung Pro drives are quite nice as well. Them and Intel are considered the top tier. I got an OCZ Vertex 4 for my laptop (using their new-at-the-time Indilinx controller) and it'd been rock solid and fast.
Jason is right, stop buying first gen drives. A modern Intel or Samsung drive will serve you quite well. It's not a bad idea to buy a size or two bigger than you need both for future proofing and for wear leveling. Make sure TRIM support is enabled as well or performance will really degrade. Actually, does FreeBSD 8 support TRIM?
-
FYI- We don't have official TRIM support in the GUI or installer, but you can enable it on 2.1
touch /root/TRIM_setAnd then reboot. It will reboot, set trim, then reboot again, and it should be active at that time.
To disable it:
touch /root/TRIM_unsetand then reboot.
-
I have a mix of Crucial M4, Samsung 830 and 840 drives in my various systems with no failures yet. Looking at the SMART logs I don't see any problems but I really should dig into these two lines a bit more to see what is going on.
173 Wear_Leveling_Count 0x0033 100 100 010 Pre-fail Always - 2 181 Non4k_Aligned_Access 0x0022 100 100 001 Old_age Always - 213 210 2 -
FYI- We don't have official TRIM support in the GUI or installer, but you can enable it on 2.1
touch /root/TRIM_setAnd then reboot. It will reboot, set trim, then reboot again, and it should be active at that time.
I just got a deal on a new Samsung 840 series for my laptop so I'm glad to hear it's a good drive. I'll look at a 32-64GB Samsung or Intel for the pfSense box.
Didn't know you had to do something to turn on TRIM - Thanks!
-
I have a mix of Crucial M4, Samsung 830 and 840 drives in my various systems with no failures yet. Looking at the SMART logs I don't see any problems but I really should dig into these two lines a bit more to see what is going on.
173 Wear_Leveling_Count 0x0033 100 100 010 Pre-fail Always - 2 181 Non4k_Aligned_Access 0x0022 100 100 001 Old_age Always - 213 210 2I've heard good things about those drives. Never used them though.
The bulk of the SSDs I use are a mix of:
-
Intel X25-E 32GB (ZFS write caching, SQL logs & tempdb)
-
Intel DC3700 400GB (ZFS write caching)
-
Intel 520 240GB 20% overprovisioned (ZFS read caching, development boxes, photo/video editing)
-
Intel 520 240GB (laptops)
-
Intel 311/313 20/24GB (pfSense w/ mSATA)
-
Transcend 8GB SLC Compact Flash (pfSense embedded boxes)
There's a couple other random one-offs floating around but these are the majority of them. I've never had a failure amongst these.
-
-
If you can get a model with a built-in capacitor (e.g. Intel 320) they have a much better chance of surviving a sudden power loss with all your data intact.
The onboard capacity keeps enough energy to flush the data to permanent storage before it goes off.
They cost more, but IMHO, it would be worth it in the long term.
Just like battery-backed RAID cache.
-
If you can get a model with a built-in capacitor (e.g. Intel 320) they have a much better chance of surviving a sudden power loss with all your data intact.
The onboard capacity keeps enough energy to flush the data to permanent storage before it goes off.
They cost more, but IMHO, it would be worth it in the long term.
Just like battery-backed RAID cache.
That, or use a UPS and configure your hardware to shutdown. Of the models I mentioned, only the DC3700 has a super cap.
-
If you can get a model with a built-in capacitor (e.g. Intel 320) they have a much better chance of surviving a sudden power loss with all your data intact.
The onboard capacity keeps enough energy to flush the data to permanent storage before it goes off.
Thanks again… just ordered a new 40GB 320 series drive. Not cheap for sure - but hopefully I'll have better luck!!!
-
Al little on the slower side but 4gb USB sticks work pretty well. I keep a duplicate for when the active one dies. And at only $5 a pop who cares if it dies.
-
The size of the drive also plays an important role in how quickly it wears out. A small ssd like a 16GB has little to no spare area to work with which increases the write amplification considerably. On all enterprise drives they have 20% or more area reserved so the drive doesn't have to reuse areas nearly as often. So a 160GB enterprise SSD might have 200GB of actual flash memory, but only 160GB is accessible by the system.
The 40GB 320 should last a long time especially if it can be trimmed on occasion.
Here is a good question. Does PFSense properly align the drive out of the box?
That can also have a very adverse affect on drive longevity if each write is split up over two block on the drive and requires both of them to be erased to put new data on the block.
-
Here is a good question. Does PFSense properly align the drive out of the box?
That can also have a very adverse affect on drive longevity if each write is split up over two block on the drive and requires both of them to be erased to put new data on the block.
Good point indeed!
-
Here is a good question. Does PFSense properly align the drive out of the box?
Is that something I could do (or check) manually?
-
Paragon Alignment Tool under Windows can show you the exact situation, and can align existing partitions (with data) to be OK for SSDs.
Linux parted creates aligned partitions for sure, but I can't remember now what options you need to enable to actually see if an existing partition is aligned or not.
Checking the first partition can be easy by looking at the numbers, but next partition numbers need to be calculated based on first partition's offsets… -
I don't think the writes are aligned, see the "Non4k_Aligned_Access" from SMART in my post above.
I thought this might be an issue and I bumped up the drive size to give lots of extra space for wear leveling to deal with it. I'm hoping 2.1 offers more SSD support options but I haven't looked at it since I don't have a spare SSD equipped box available right now.
-
I don't think the writes are aligned, see the "Non4k_Aligned_Access" from SMART in my post above.
I thought this might be an issue and I bumped up the drive size to give lots of extra space for wear leveling to deal with it. I'm hoping 2.1 offers more SSD support options but I haven't looked at it since I don't have a spare SSD equipped box available right now.
2.1 won't (aside from trim I mentioned above) but 2.2 may.
-
Samsung has a new batch Enterprise ready drives for folks that have systems that are going to really beat on it, 3.4 PB of writes.
http://www.theregister.co.uk/2013/05/23/samsung_s843t_upgrade/
I'm really partial to the Samsung drives, not for any hard reason but because I like the idea that they make their own memory chips and controllers and don't rely on outside vendors for them.
Jimp, 2.2… Waaahhh! I haven't even started whining "are we there yet" over 2.1 and now I'm waiting on 2.2. :-)
