It's worth to use an ssd instead?
What are you hoping to gain?
If not, I read about f2fs or something similar. It was about a file system made for USB key. Should I format my USB key in that file system?
I don't think the grub included with Debian could boot off it. It isn't going to magically give your usb key more writes either.
The only potential benefits I can see are io and lifespan.
But if you are looking for a cheap SSD, chances are you will not be buying a good brand, and then you must ask, how much more do you trust an ssd that costs what a decent thumb drive costs?
Just my two cents. Unless you were willing to shell out for a decent brand of SSD, I would not bother.
But if you are looking for a cheap SSD, chances are you will not be buying a good brand, and then you must ask, how much more do you trust an ssd that costs what a decent thumb drive costs?
I have bought quite a few used 16GD Samsung SSDs for less than $20US shipped from ebay. There are still some available as well as other brand names in that price range. My OMV has been running on one of them for more than a year without any problems.
The only potential benefits I can see are io
I/O increase on the OS drive doesn't benefit much.
I/O increase on the OS drive doesn't benefit much.
Fair enough, I just meant strictly speaking, those are the only two benefits I could think of- throughput and reliability. Not necessarily that it will make anything faster, just that the throughput of a thumb drive is peanuts compared to an SSD.
It's worth to use an ssd instead? The cheapest one cost around 45€.
Nope
Maybe you should consider they type of flash memory you buy. It's probably more important than what it's in. There are several types.
SLC Generally the highest endurance ratings of the lot. Each storage cell just holds a 0 or a 1. There is some mention of this type in the OMV doc's. Those mention looking for parts that also have wear levelling. It's expensive stuff but bargains do crop up.
TLC This measures voltages in each cell so can hold more values than just 0 or 1. Done because parts cost is largely down to the area of silicone they use.
MLC Not so clear what is going on in these. No doubt it's cheaper to make than others and initially there were moans about it on the web - oh no not more voltage levels so bound to be even worse than TLC. I've seen reliability reports that suggest they can be or are significantly better than TLC but still no where near as good as SLC.
We might see parts called eMLC at some point. Better still but well short of what SLC can achieve.
Make might matter on SSD's. I've seen suggestions that Intel ones will all fall over when they reach their stated data ratings. Corsair on the other hand went for a lot longer but didn't all fail at the same time.
Writing to them is a major failing feature. I believe there is an omv plugin that can help with that. Data retention times is another - not helped by using voltage levels so SLC is likely to be much better. I found some 2gb slc usb sticks. Hopefully these will be better than the usual usb stick but would they be up to the standard of the slc caches fitted to HP servers? Some how I doubt it.
I run an HP Xeon workstation as my main PC. I've had a Plexor ssd boot disk in it since 2013. What I have done though is mounted /var. /tmp and /home to disks. That is a TLC chip. Way to big for just software.really. I've no idea how much scope there is for this sort of thing on OMV. Var is the biggest culprit for system writes pus tmp of course. Swap can be too but I don't mount it on my main pc. No point as I don't use suspend. Some people mount /var/log to ram I decided to mount the lot to disk.
In terms of limiting writes I've seen suggestions that using EXT2 may be a good idea. Even EXt4 increases the write load and some others really do. On ssd's in particular ext2 may make a lot of sense. All sorts of things are going on inside them. I have a feeling that I will be doing this on OMV and adding a cheap UPS.
Out of interest I run wd reds in mirrored raid using ext4 on my main PC. I don't check it that often but some one turned the supply off while I was using it. Result mdadm dropped one of the disks. I can't be 100% sure but it looks like that is what happened. Better at spotting problems than preventing them. On that basis a UPS makes more sense.
John
-
with regard to SLC usb keys, I found a pretty reliable source, I believe...
see here, it is the vtran line of SLC USB keys, he has a little explanation about the differences between keys. They ship from Hong Kong and ems is available.
Here is a quote about the types of memory...
- Today, manufactories are using 3 kinds of flash chip SLC, MLC and TLC to build USB drive and SSD drive.
- SLC (single layer cell), each cell store only 1 bit data 0 and 1. using two voltage VL and VH for 0 and 1 . MLC (multiple layer cell), each cell store 2 bit of data, 00, 01, 10 and 11. Using 4 different voltage to represent 2 bit of data TLC (triple layer cell), each cell store 3 bit of data 000,001,010,011,100,101,110,111. Using 8 different voltage to represent 3 bit of data.
- Compare the speed of three different Flash, the read speed is the similar. For MLC, the write speed is 1/2 of the read speed. For TLC, the write speed is 1/3 of read speed.
- Some of the manufacture, using TLC for their product, but program part of the TLC in SLC mode, to improve the testing figure.
- For the life time, SLC have 100,000PE (program erase cycle), MLC 3000-10000PE, TLC 500-1000PE.
- Becasue of high density, the price of SLC Flash is expensive.
And I found this review on chiphell which makes me think they are pretty legit, you will have to use google translate but you can get the gist of what they are saying.
thanks a lot to everyone! As someone already guessed I was looking for the lifespan of a usb key.
My concearn is that one day it will simply stop work 
SSD are not different, but they have so many R/W that I'm not worried even if I won't change it in 10 years!
The chepest that I found was a 60GB corsair for 45€. Now I looked for a kingdian SSD and I found one for 20€. I don't care about performance, I care about lifespan and stability of my system.
There is any way to monitor my usb key health or if it's slc/mlc/tlc?
The other catch with flash is data retention time. The cells store a voltage and there will be some leakage over time - something they are for ever working on. So again slc wins as it's only storing levels representing 0 or 1 so more voltage drift is acceptable than those that use several voltages to represent more numbers in a single cell.
I'm going to start using flash for my /home directory so that's why I'm going for an OMV set up. I'll need to back up my data. I bought a couple of these
http://www.ebay.co.uk/itm/ATP-…ksid=p2057872.m2749.l2649
Good? Don't know so time will tell. The spec's do say the correct things though. King Diane spec's look good too but they are very cheap if they can achieve the figures they mention on them so probably can't but the only way to find out would be to try one.
John
-
@Blabla Not really. I would also take most of this with a grain of salt.
For one, most of the mainstream usb drives I researched on the market (ie Kingston, Corsair, Transcend etc) seem to be using MLC. I have a few SSD which are also MLC.
If you want to monitor the health then you should look for an SSD which supports S.M.A.R.T. I think!
The other catch with flash is data retention time.
Irrelevant for this use case since the drive is powered all the time anyway.
Wrt http://www.ebay.co.uk/itm/ATP-…ksid=p2057872.m2749.l2649 the only interesting thing there is 'TBW** (max.) : 96 TB'. So the seller tells you that this stick with 2GB capacity can be fully overwritten 50,000 times (96 * 1024 / 2 = 49152). If you fear about flash storage wearing out too fast your next step is starting to study 'write amplification' behaviour since it matters whether you try to (over)write 1 bit at a time or eg. 4096 bytes (write amplification can differ by a factor of up to 4096 then -- in other words: when writing always only 1 bit of data your flash might wear out 4096 times faster).
When you care about write amplification you start to tweak settings (higher commit interval, folder2ram plugin) since this will reduce write amplification drastically. If you care about how Wear Leveling works you simply buy higher capacities (if 4GB are needed, buy 64 GB since your flash will last 16 times longer). If all of this does make no sense to you and you still fear flash media dying then simply buy a good SSD from those vendors who report 'lifetime expectancy' through S.M.A.R.T.: https://unix.stackexchange.com…or-the-mediums-wear-level (so always check first which vendors do provide this and then also how to interpret the values).
I pulled recently a Crucial SSD out of a customer's MacBook Pro after 6 years of heavy use (Photoshop). According to SMART the SSD was at 93% of it's lifetime expectancy (7% worn out) so in other words with the same usage pattern it will be good for another +90 years before it will finally wear out. On all storage servers we monitor iostat output daily. On none of the systems the amount of data written to the thumb drive containing the rootfs exceeded the drive's capacity (so even with pretty worse write amplification these drives will not wear out the next decades)
I personally stopped thinking about flash cell types and such stuff since the main problem these days is counterfeit flash media. You buy something from a renowned vendor that is advertised being SLC while in reality it's cheap crap having only a fraction of the advertised capacity. Some of these fake media can be detected by running the usual suspects (F3/H2testw to check for fake capacity -- common with both USB sticks and SD cards) but the better fakes do have the real capacity, behave rather well when tested with sequential workloads (as the above tests) but start to show their real nature only after some heavy use.
Since this turned from a technical into a market problem (fraudsters being encouraged to counterfeit storage) when it's about USB thumb drives I personally try to buy brands that have their own NAND flash fabs and also create retail products (to minimize the risks involved when 3rd parties assemble USB thumb drives or other flash media). And I only buy through channels where I can return stuff without questions being asked since dealt way too often with counterfeit crap already.
Irrelevant for this use case since the drive is powered all the time anyway.
The cells are essentially capacitors which is why they can store a specific voltage and is why that have this problem, powered or unpowered. One thing that can be said is that figures are often quoted long before the parts have been out long enough to establish the duration so they are generally predictions based on measurements and assumption. Much depends on the quality of the fabrication process and in practice retention times are likely to be longer. Maybe they have found a solution but I don't think so.
Your point about fakes is an interesting one. Say a manufacturers fab isn't all it should be. It might make sense to produce slc which wont be up to the spec of say Intel's but is still better than other types. How much better is debatable and if really bad might even be worse. SLC will be the same as other types. There will be both good stuff and bad stuff about. There are Chinese companies that have their own fab. You might say that's the problem. It can be pretty dismal to just as good as the major players. I have some electronics parts from a large chinese manufacturer for my own use. All I can say is that at least on these they are honest about the spec but they are way way worse than the same part from the usual manufacturers. Ok for me though. Essentially wondering about the actual source is a problem on standard electronics components these days. For my own use I try and buy of main suppliers - the same ones as I would use at work but there I had the advantage that lots came directly from the actual manufacturer.
Buying a higher capacity ssd than is needed to extend write life isn't as simple as that. It's what I did on my PC. It's a 120gb part, way to much for just software that gets updated now and again. My software installation is about 14gb. Last week I thought now is maybe the time to trim it and found that the system is doing this automatically. Probably a feature of ext4 so to maximise life I have to hope that the ssd does have it's own wear levelling built in. It seems to have it. Maybe all have. After 4 years I feel it's time to change it anyway and will be next time my distro updates. Things move on in 4 years.
I'm going to move to all flash on my pc anyway but will choose the make and source carefully. I can experiment a little with omv. Hard drives these days aren't what they were. I run a xeon hp workstation. It came with one disk. People wouldn't believe the smart results on that. It shows errors corrected via parity as well as errors that would cause problems. None of those but an unbelievable number of read parity corrections.
John
-
There are Chinese companies that have their own fab. You might say that's the problem.
No, not related. I was talking about what usually happens where consumer flash products are assembled. Old story but still worth a read: https://www.bunniestudios.com/blog/?page_id=1022
That's why we only buy Samsung and SanDisk currently if it's about USB thumb drives or SD cards (since those are amongst the few manufacturers out there that are at least able to control the whole production process using products from their fabs and combining it with their own knowledge and even controllers).
But even then you can happen to get fake cards, a simple web search for 'samsung evo counterfeit amazon' or something like that should give the idea (you get fakes EVERYWHERE since resellers can't avoid getting this stuff inserted into the supply chain somewhere. That's why I focus on sellers that try to avoid this happening and at least don't discuss when you ask for a return/refund if you found something suspicious when testing a new USB thumb drive).
Wrt SSDs I would never buy some cheap no-name stuff but only from companies having their own fabs and expertise (I believe we only use Crucial/Micron, Intel and Samsung in the meantime, at least their SSD all support a SMART attribute exposing the 'wear out' percentage so you can act on accordingly). With those SSDs it's also easy to check for counterfeits (I don't believe something like this exists yet) since you can benchmark them and even check the serial number if in doubt. 'You get what you pay for' in this case 
But I would not use a SSD for boot storage, just a medium sized quality USB thumb drive with flashmemory plugin enabled. We bought recently a bunch of 'SanDisk Ultra Fit V2 32GB' for 15 bucks each. Sequential performance ok, random IO not that great but that's irrelevant for this use case anyway (same with filesystems -- I was curious whether there are obvious F2FS performance advantages over ext4 on flash media but I don't think so after some extensive testing). The amount of data written to these thumb drives through their entire lifetime will most probably not exceed even 1TB so there's no reason to fear wear out at all.
BTW: All consumer oriented flash storage implements wear-leveling (even the cheapest ones) otherwise they would be worn out way too fast with usual filesystems like FAT/FAT32/exFAT. The cheap devices implement only a primitive wear-leveling while good SSDs shine here since they're doing wear-leveling in a way that sustained storage performance remains high even if the media is used with +95% (then you'll see the difference easily between primitive and advanced implementations and controllers)
It came with one disk. People wouldn't believe the smart results on that.
If this was a Seagate then most probably everything is fine since you're not talking about 'smart results' but in reality 'smart raw data' that needs correct interpretation: http://www.tomshardware.co.uk/…art-values-seagate-normal
There are only a few SMART attributes we as end users can make use of ('lifetime expectancy' on good SSDs being one of the best examples) and more generally speaking I would recommend to focus on relevant stuff. For me that means 'data integrity' that's why we only use btrfs/ZFS for valuable data. Works even with a rootfs on an USB thumb drive and provides the ability to recover from failed updates and stuff like that (see here for an example)
Looking around I do need to update that post a bit. Retention is worse when powered down. Some of the numbers mentions are not at all nice especially as the temperature goes up. The problem when they are powered up should be managed by the chip - seems it moves stuff around before it becomes a problem.
If people are interested a couple of pages
https://en.wikipedia.org/wiki/Flash_memory
http://www.flashstorage.com/fl…echnical-economic-primer/
Wouldn't it nice if all producers provided full info on what they do - no chance on consumer stuff. However it might be worth looking for it just in case it is there somewhere.
I found a comment for people running an IBM data centre using flash - never power it down for more than 40 days and only if you must. Best leave it on 24 / 7 as it doesn't use much power anyway. No idea how old the info is. Seems it also gets worse as the device ages.
It can be hard to find truly up to date info on subjects like this. I've used another sort of flash, E^2 professionally. Initially that had low retention limits - it crept up to 100 years on some but effectively that is a form of slc. As the stuff hasn't been around that long it can only be a prediction. It was -40 +85C stuff too. Not 0-70.
John
-
It came with one disk. People wouldn't believe the smart results on that
Why/how should this be an issue with this use case here? An USB thumb drive will loose data eventually but not when powered all the time as a boot disk. And 'archiving' on normal flash media is almost stupid given the price difference to HDDs.
I found a comment for people running an IBM data centre using flash
There's no magic involved but flash storage in data centers is usually something different than what we're talking here about (if you buy cheap USB thumb drives today they might be made of recycled NAND flash ripped out of old smartphones with an USB controller masking the bad areas).
Usually in data centers we're using tired storage with the fastest layer made of flash (expensive stuff with very low latency and high IOPS numbers even with low queue depths), then 1 or 2 layers performant SAS/SATA, then maybe another layer with cheap 'nearline storage' (cheap SATA stuff with high capacity) and then in the past sometimes even tape robots (storage density is still good especially when comparing with data retention numbers). But if the use case is all about lowest latency and high random IO performance only relying on flash can be an option. But then we're talking about Enterprise SSDs not even relying on SAS/SATA since for modern multi-core servers with many requests in parallel you also need protocol implementations that can cope with that (NVMe) to get decent performance.
The comments from IBM are pretty clearly ancient.
Enterprise ssd's are high quality slc's and I doubt if many data centres use them exclusively.
The disk I mentioned. It uses similar fixes to those used in ssd's to get round problems - parity to get round storage mishaps. It's the only disk I have come across that provides the info via smartctl. The increasing density on magnetic disks has problems all of it's own.
Any way the OP now has sufficient info to answer the question they asked, Who they choose to believe is up to them. There isn't a straight answer to the question really other that in some ways it's a case of buyer beware. There are unlikely to be any extreme near free lunches in this sort of area unless it's superseded stock that is being disposed of. These days releases are usually staged to get round that problem or a very high price penalty is attached to the latest greatest especially if it really is better than the part it will eventually replace.
If some one wants to nose around a true flash fabricator this site might be of interest. Note the logo's at the bottom of the page.
https://www.micron.com/product…h/choosing-the-right-nand
John
-
- As someone already guessed I was looking for the lifespan of a usb key.
- My concearn is that one day it will simply stop work
The chepest that I found was a 60GB corsair for 45€. Now I looked for a kingdian SSD and I found one for 20€. I don't care about performance, I care about lifespan and stability of my system.
As ryecarron noted, boot drive I/O on a server platform is not important AFTER the boot process is complete. (And I reboot once a week, around midnight, where the speed of the boot doesn't matter.) And that's not saying my boot time is slow either. With a San-Disk USB 3.0 drive, my boot time is less than a minute.
______________________________________________________
Beyond that - the bottom line in this is,,,, all forms of media fail. It's not a question of "if", it's "when" and no one can tell you with any certainty how long even name brand media will last. (Warranties don't do it for me either. Simply refunding the purchase price or providing a blank replacement doesn't help with the loss.)
With that noted, it absolutely makes sense to prepare for the inevitable failure with backup. Since they're cheap, why not use a USB 3.0 thumb drive as an OMV boot drive? With the flash memory plug in, I've found that flash media lasts a long time. Given how cheap they are, and the ease of cloning USB sticks with a utility like Win32 diskimager (and others), it makes sense to have an extra copy (or two) of the boot drive ready in the event of a failure. Lastly, with USB connections on the outside of the box, it's dirt simple to replace the boot drive without even turning a screw driver. Even with a complete boot drive failure, with clones standing by, you can be up and running in a minute or two. And if something goes wrong with an upgrade, you can back out of it just as fast.
There is any way to monitor my usb key health or if it's slc/mlc/tlc?
USB or SD Cards don't support SMART yet, so I believe the answer to "monitoring" is "no". The best that can be done would be a utility like H2testw, which works something like a crude version of Spinrite. After a fresh reformat, the utility destructively writes known files and reads them to determine if there are errors in flash media. If errors are discovered, it's best to just toss the stick.
Why rebooting anyway? I haven't rebootet my NAS since almost half a year. (Switched to OMV3) Before that it was running OMV2 almost 2 years without reboot. So why bothering how fast the OS drive is.
Performance becomes important when frequently used data is stored on the OS drive. I think of databases from MySQL for example, which are not stored on data drives. And energy consumption is a point to think of. For USB thumbdrive, wear leveling and memory cell type is another fact to keep in mind when used as OS drive. (Flashmemory-plugin) I chose a SLC USB thumbdrive ...
Sie haben noch kein Benutzerkonto auf unserer Seite? Registrieren Sie sich kostenlos und nehmen Sie an unserer Community teil!
