|
Hello guys, I don't know if perhaps TL is the wrong place to post a question like this, but historically there have been very resourceful people on here who give better advice than people on tech-dedicated forums. First off, I want to say that I'm not very savvy as far as networks go. I can build computers and I understand some basics but I can't be bothered to learn to operate Linux. As such, it would be preferable for the solution to be relatively simple.
I'm thinking of building a RAID5 NAS for my storage needs and I was going for 4x 3TB drives which would give me 9TB of storage and 3TB for redundancy. It needs to be able to somewhat quickly transfer files, at least around 40-50 megabytes per second, and it needs to be able to stream 1080p movies through a gigabit router.
For a while I considered the possibility of using FreeNAS and a custom-made computer, but after emulating it for a while, I came to the conclusion that it's a bit complicated and I'd just end up deleting my data trying to make minor modifications to the array or something. Also it almost seems like you need to "mod" the OS so that it'll automatically warn you when a drive dies...
Lately I've been looking at QNAP 4-bay units which have convenient hotswappable bays, but if I picked that up, I wouldn't know what hard drives to get. Plus, performance CAN be an issue because NAS units usually have cheap hardware.
Are there any other decent options that I would look at? My budget is around $900-1000 for the NAS and its hard drives, which is somewhat modest but it's hopefully sufficient.
Opinions/Ideas?
|
I actually went through all these steps and figured out that making a Raid array makes 0 sense.
The advantage of a Raid array is in load balancing: it lets you access any file with a fixed performance hit. Say in your typical non-raid setup, there are two files on two hard drives. When you access two files, there is a chance that it is on different hard drives and a chance that they are on the same hard drive. With Raid, the files take the same amount of time to load. Without raid, they may take different times to load.
This doesn't matter to a casual user. In fact, I found that just having the extra hard drives let me put those big library files on different hard drives, then symmlink them all to the same directory, and i got a boost equivalent to having a $1000 raid card.
Worse, RAID is actually not designed to protect your files. The best way to protect your files is redundancy, but RAID does not provide redundancy because there is a single point of failure: the actual raid controller. Raid controllers are not interchangeable, and aren't designed to be interchangeable. RAID performs worse on large hard drives too, because it's common for a entire hard drive to have some small corruption of files. The problem here becomes that if you lose only one file on one disk, and a disk goes bad, best case is you lose corresponding files on the other four disks: worst case is you lose all the data on all the disks. So data loss is actually magnified by RAID. So having a raid is in fact providing less protection as having all your data on one hard drive. What you really want to do is look into some software pairty generating software. This way you can backup the data from 3 drives onto 2 drives, so you can deal with two failed disks. This is the minimum amount of data redundancy you should do with 3 gb hard drives.
|
The good thing about RAID is its automated nature, but your concerns are valid. Still, if the controller died, you can always find one that's compatible regardless of the fact that not all of them are.
|
On June 07 2013 03:22 GoldenH wrote: I actually went through all these steps and figured out that making a Raid array makes 0 sense.
The advantage of a Raid array is in load balancing: it lets you access any file with a fixed performance hit. Say in your typical non-raid setup, there are two files on two hard drives. When you access two files, there is a chance that it is on different hard drives and a chance that they are on the same hard drive. With Raid, the files take the same amount of time to load. Without raid, they may take different times to load.
This doesn't matter to a casual user. In fact, I found that just having the extra hard drives let me put those big library files on different hard drives, then symmlink them all to the same directory, and i got a boost equivalent to having a $1000 raid card.
Worse, RAID is actually not designed to protect your files. The best way to protect your files is redundancy, but RAID does not provide redundancy because there is a single point of failure: the actual raid controller. Raid controllers are not interchangeable, and aren't designed to be interchangeable. RAID performs worse on large hard drives too, because it's common for a entire hard drive to have some small corruption of files. The problem here becomes that if you lose only one file on one disk, and a disk goes bad, best case is you lose corresponding files on the other four disks: worst case is you lose all the data on all the disks. So data loss is actually magnified by RAID. So having a raid is in fact providing less protection as having all your data on one hard drive. What you really want to do is look into some software pairty generating software. This way you can backup the data from 3 drives onto 2 drives, so you can deal with two failed disks. This is the minimum amount of data redundancy you should do with 3 gb hard drives.
I really really really hope that this is just a troll.
|
This would require a bit of research, which I could do and offer some advice later.
However to begin with if you are wanting that much data I would stay away from RAID5. As disks keep increasing in size RAID5 becomes less efficient. When a single hard drive in a RAID5 Array dies, you still have 100% of your data. That is the great thing about Raid 5, the downside is that when a hard drive dies you must allow your RAID to be in a degraded state as the failed hard drive is replaced and RAID array has to be rebuilt. With large hard drives this can take very very long time, more importantly as disk sizes increases so is the likely hood the repair will fail, leaving you will a degraded non redundant RAID array.
I would recommend looking into RAID 10 (1 + 0). This uses striped and mirrored into one, creating redundancy and easy restoration. If a hard drive dies no actual array needs to be rebuilt, only a straight copy of data over to the new hard drive. This is the preferred method with large disk drives.
I would begin looking at NAS devices made by these companies:
http://www.synology.com/us/index.php
http://www.thecus.com/
|
Well the data is rated at "would suck if I lost it" but it's not of critical importance, so I'd be sort of comfortable with RAID5. The problem with 1+0 is that it's quite expensive, I lose half my drive's space. I'd end up with 6TB instead of 9TB, which would be filled up within a few years. I could RAID10 with 4x 4TB for a total of 8TB but now I'm looking at well over $1000.
I do understand that a danger of RAID5 is the long rebuilding time during which there's a chance that another drive could die, but I didn't know it could just outright fail to rebuild. :o
As for the companies, I'll look into thecus, although I'm worried about their support in NA. And Synology's equipment seems to be very expensive.
|
Please do not use RAID5, it is effectively deprecated with 2TB+ disks being all too common. The chances of another disk failing during the rebuild that follows one dead disk is just too high. Sorry but I can't find the artcile I usually cite for this.
You should try looking into building a ZFS box, it's softRAID, yes, but highly reliable.
|
On June 07 2013 03:58 Weiman wrote: Please do not use RAID5, it is effectively deprecated with 2TB+ disks being all too common. The chances of another disk failing during the rebuild that follows one dead disk is just too high. Sorry but I can't find the artcile I usually cite for this.
You should try looking into building a ZFS box, it's softRAID, yes, but highly reliable. Maybe I should change my plans then... As for ZFS boxes, not only does it looks complicated, it also seems to be very expensive because apparently ZFS is very dependent on ram... It's not looking good for my NAS plans, I guess
|
Currently this 4TB hard drive is $150 (on sale)
The Synology Disk Station DS413J NAS runs around $400
That is roughly $1000 before tax.
Synology states that device will get 50MB/s up and 100MB/s down, but they test with SSD's in RAID 0.
Of course that is just one option, I think you should be able to get what you want within your budget, or maybe just slightly above.
I would opt for Raid 10, and keep your eyes open for deals on 4TB drives.
|
On June 07 2013 03:22 GoldenH wrote: I actually went through all these steps and figured out that making a Raid array makes 0 sense.
The advantage of a Raid array is in load balancing: it lets you access any file with a fixed performance hit. Say in your typical non-raid setup, there are two files on two hard drives. When you access two files, there is a chance that it is on different hard drives and a chance that they are on the same hard drive. With Raid, the files take the same amount of time to load. Without raid, they may take different times to load.
This doesn't matter to a casual user. In fact, I found that just having the extra hard drives let me put those big library files on different hard drives, then symmlink them all to the same directory, and i got a boost equivalent to having a $1000 raid card.
Worse, RAID is actually not designed to protect your files. The best way to protect your files is redundancy, but RAID does not provide redundancy because there is a single point of failure: the actual raid controller. Raid controllers are not interchangeable, and aren't designed to be interchangeable. RAID performs worse on large hard drives too, because it's common for a entire hard drive to have some small corruption of files. The problem here becomes that if you lose only one file on one disk, and a disk goes bad, best case is you lose corresponding files on the other four disks: worst case is you lose all the data on all the disks. So data loss is actually magnified by RAID. So having a raid is in fact providing less protection as having all your data on one hard drive. What you really want to do is look into some software pairty generating software. This way you can backup the data from 3 drives onto 2 drives, so you can deal with two failed disks. This is the minimum amount of data redundancy you should do with 3 gb hard drives.
There's so much wrong in this post... just quoting it to mention that.
Anyway, @TS: RAID-5 is not really that interesting in the era of large harddisks, as Weiman pointed out already. The main reason for this is the UBER (which is just too cool of an acronym not to use), or Uncorrectable Bit Error Rate. This is the rate at which the disk will spit out an error in a single bit, averaged over a long period. Most consumer-level disks are rated at 10^-14, which means 1 in every 100,000,000,000,000 bits will be wrong, on average. Looks impressive, but it boils down to an error every 12.5 TB of data access on average. If you have a 4x3 array in RAID-5 and one disk fails, you need to read 9 TB of data to regenerate the array with a replacement disk. Chance of hitting an error is pretty decent, at which point data loss occurs and the whole point of RAID-5 is a bit gone.
To remedy this, we could just increase the raid-level (because bigger numbers are better, obviously) and use RAID-6, which works the same as RAID-5 except that 2 disks are used for parity, so you can suffer 2 disks failing without losing data. But your net storage volume will be lower as well.
Since RAID doesn't offer any form of reliable backup-mechanism (it only protects against disk failure, which is only one of many ways you could lose your data), its only real advantage is that it allows the system to stay operations when a disk fails. This can be crucial in a business, where downtime means loss of productivity, but for a home user it doesn't really add that much. If your data is non-critical (that is, it's not even worth backing up), I would just use the disks separately or set up some form of disk-spanning (where the disks are combined into a single volume but without any redundancy features).
An excellent alternative is a ZFS-based system. ZFS is a newish filesystem that has a couple of cool features. Primarily, it stores a checksum for every data block. This checksum is stored on a physically different location, so any bit errors in the data can be detected. Furthermore, ZFS allows you to use RAID-like setups, with RAID-Z1 being the most interesting. This setup is very similar to RAID-5, distributed parity using up the size of a single disk. When you combine the error-detection capabilities from the checksumming with the data redundancy from the RAID-setup, it becomes possible to detect bit error and repair them. A RAID-Z1 array is not as susceptible to the same UBER-problem that I mentioned earlier, since it can detect errors rather than simply accepting every bit the harddisk reads as correct.
You can do more cool stuff. Suppose you have some data that you would like to protect, but most of it is throwaway. You can make a 200 GB partition on each of the 4 disks and create a RAID-Z1 array with those (or RAID-Z2 for the RAID-6 equivalent), giving you 600 GB of storage with checksumming and redundancy. The remaining 2.8 TB on each disk you just add to a span or even a stripe (RAID-0), giving you 11.2 TB of high-performance storage with no redundancy (though checksums are still made, so the system can tell when data is damaged). Since ZFS is a purely software based solution, you can move the harddisks to a different computer and reuse the array without data-loss.
ZFS is available on FreeNAS, but it also has stable implementations on some Linux distributions. Unfortunately, it is not as plug-and-play as most of the pre-built NAS devices, but if you put some time into setting it up, you get a very robust system using a very modern filesystem.
tldr: No point in RAID for consumer-purposes unless you do it with ZFS.
edit: Noticed your comment on RAM & ZFS-performance. It's true that ZFS likes RAM to boost performance (ideally as much as 1 GB RAM per 1 TB storage), but the question is how much performance you really need. I built a low-end ZFS machine with just 4 GB RAM and a very weak CPU and it pushes files over the network at 75 MB/s, which is not that far from the cap of a regular gigabit ethernet connection. Not that many pre-built NAS devices beat that performance. So don't worry about the RAM-needs of ZFS, the recommendation is mostly true for very large systems in corporate settings.
|
I guess I'll keep looking around. It's so strange that they would test all NAS'es in RAID0, who buys a NAS for RAID0? And SSDs, come on...
And yeah ZFS has interesting uses but I don't trust myself to handle it properly. Thanks for the post Rannasha, that clears things up!
|
|
|
|
|
5930 Posts
On June 07 2013 04:38 Djzapz wrote: I guess I'll keep looking around. It's so strange that they would test all NAS'es in RAID0, who buys a NAS for RAID0? And SSDs, come on...
And yeah ZFS has interesting uses but I don't trust myself to handle it properly. Thanks for the post Rannasha, that clears things up!
Small Net Builder should test RAID 5 and RAID 1 performance with the Intel NAS Performance Toolkit, since that's what people actually use.
For most things, Synology and QNAP consumer options both offer perfectly usable NAS systems. When people/benchmarks say its slow, they mean it can be slow at rebuilding RAID arrays and dealing with encrypted data (this you probably do not care about). For a typical home user who simply wants to centralise and protect their data, they're actually perfectly usable. Just limit yourself to content consumption and those NAS systems won't perform exceptionally poorly.
If you actually want to try Synology's OS on a computer, Xpenology's forums have instructions on how to get it working on a DIY computer. You just need a USB stick and a spare hard drive. Here are instructions. Just make sure you follow the instructions properly, you need to be quick about removing and installing the USB stick.
If you ever decide to go the DIY route, I strongly suggest the HP ProLiant MicroServer N40/N54 (~$200-$300) because its a nice form factor, its dirt cheap, the performance is very good, and it has PCI slots so you can add better network cards. Unless you have an entire computer lying around or need a system that can hold more than 5 drives, there's no reason to actually build your own system. And yeah, I'd agree that FreeNAS suffers from the problem most open source projects suffer from: usability for beginners.
|
Thanks Wom. I think I would have to limit myself to consumer options. The Synology OS seems like an interesting option but I don't know that I want to be messing with mods... on the other hand the installation looks simple enough and the guy says he used it with a HP N40L which you mentioned and it does in fact look like a good way to have performance and good features. But it doesn't seem to come for $200-300 here.
That said, do you agree that for the size of what I'm trying to do, RAID5 is a bad idea? Should I do RAID10 instead?
|
5930 Posts
In the past, HP sold the N40 with a cashback but I'm not sure if they currently do this. When I was flirting with the idea, I got mine for around $250 Australia dollars from eBay. It was a grey import from Taiwan but that wasn't much a concern since it was less of a server and more of a toy for me. It even came with 4GB of ECC DDR3 memory, which is kind of cool.
RAID 5, RAID 1, or a vendor's version of RAID (like Synology's Hybrid RAID) should be fine for you as more than 1 disk shouldn't die simultaneously*. If you need to transfer files "at least around 40-50 megabytes per second" and watch "HD video", most decent low end NAS systems should be able to do this without any hassle. In normal cases, something mid-range like a Synology DS413 should do read and write at around 80-100MB/s in RAID 5.
RAID 10 is neat but RAID 0 with hard disks is seriously overrated and you're just flushing money down the toilet by doing this because you're basically halving your available data storage by half. Look at it this way, does speed matter if you don't notice it?
* Don't buy drives from Newegg until they learn how to package them properly. I think a stupidly high percentage of WD Red drive failures can be sourced to Newegg.
**I suggested the Xpenology idea so you could play around with the Synology software. Its good to get an idea of what it can actually do before buying it.
|
Alright! And no worried, I try to avoid doing business with Newegg whenever I can. I like NCIX and recently I've noticed that Memoryexpress's pricematching is nice and its $5 flat rate shipping is pretty great.
On June 07 2013 23:28 Womwomwom wrote: RAID 10 is neat but RAID 0 with hard disks is seriously overrated and you're just flushing money down the toilet by doing this because you're basically halving your available data storage by half. Look at it this way, does speed matter if you don't notice it? I don't care for RAID0, but you don't lose half your storage with it. Did you mean RAID1? Mirroring is kind of nice, I have 2 mirrored WD greens and it provides a good amount of peace of mind. It doesn't constitute a backup by any stretch of the imagination but it feels relatively secure.
So many people seem to be opposed to RAID5 with 3TB drives though, it's kind of scary to try.
|
5930 Posts
RAID 5's major problems with bit rot and UREs are manageable if you do regular data scrubbing. I can see the justification to choose RAID 10, due to bit rot in systems that aren't utilised and UREs, but that's kind of a decision you should make yourself. I'm not surprised it gives small businesses and enterprise the shakes, though.
Thinking about it again, I probably wouldn't put 4x 3TB drives in a RAID 5 array either, especially if they're consumer grade drives. You're asking for drive failure and rebuilding the damn thing will take a year and the amount of data you need to rebuild is asking for UREs.
RAID 10 has the advantage of using mirroring but the performance boost you get from it probably won't be really noticeable and the number of drives you need is quite a lot. I would just use RAID 1 if you like the mirrored route, I don't think the performance hit will be seriously noticeable if you're not expecting a lot of performance from your system (40MB/s isn't actually that much, everything should be able to do that in RAID 1).
Traditional RAID arrays are kind of unwieldy to be honest, which was why I was never a huge fan of the RAID 5/6. That's kind of why NAS systems often come with vendor RAID software that seem to work like Windows Home Server's drive extender. GoldenH's SnapRaid suggestion is actually perfectly valid because there's a lot of third party applications designed to make the process less confusing - I used to use Windows Home Server with Drive Bender.
|
I run a Synology DS411j with 4x2TB drives in RAID 5 and have had only one hiccup in two years. It was during a move that afterwards left one of the drives to fail on powerup. I ran a deep SMART test and it came up ok, so I rebuilt the array and it has been working fine since (seven months). The rebuild took 20+ hours though, with about 3TB worth of data.
I have no qualms with putting 4x3TB drives in RAID 5 as long as one does regulars scrubbs. I do it monthly and run SMART tests once a week for peace of mind.
It is just for storage though. Anyone using a NAS for backup are fooling themselves. Use a spare drive for the really important stuff and keep it at a seperat location, use a free cloud based sollution like Dropbox if it is not huge amounts of data or shell out the cash to buy more if you need it.
|
On June 08 2013 02:34 Hokum wrote: It is just for storage though. Anyone using a NAS for backup are fooling themselves. Use a spare drive for the really important stuff and keep it at a seperat location, use a free cloud based sollution like Dropbox if it is not huge amounts of data or shell out the cash to buy more if you need it.
Using a NAS for backup is fine as long as you know what you're doing. I have my important data on my desktop (working copy) with a backup on my NAS/server which has a RAID-Z1 array and a second backup on an external disk in a locker in my office that I take home from time to time.
I trust my NAS with ZFS far more when it comes to my data than the single NTFS disk in my desktop.
|
|
|
|
|
|