to RAID or not to RAID?

[Shifte]

Hi guys!

Long time lurker looking for some advice. I am in need of building a home media server. I currently expect to need a minimum of 4TB of storage but would like at least 8TB for room to grow. I plan on purchasing all hard drives at once and am thinking 2TB drives would be a good fit.

I am storing mainly movies (1.5TB) and tv shows (1.3TB) that i have ripped/recorded. Some of the Movie DVDs/Blu are damaged an may/may not be able to be reripped in the future. Some of the TV shows are no longer on air and would require me to purchase the DVD/Blu should i lose the rips. I know that data redundancy is important but am unsure what route to go. I also plan to back up my laptops and desktops to the file server.

I will be using an old Core2duo i have lying around in a mATX case to cut down on space. i will need to purchase RAM, Power Supply and Operating System. I will need remote access from 2 tablets, 2 android phones, 2 windows 7 laptops 1 Windos 7 desktop , maybe a streaming device such as a AppleTV/Roku and possibly a HTPC in the future.

Some things to consider: Cost isn't an issue to a degree, I have the money to spend and want the best value.

QUESTIONS:

1. What/if any RAID would be the best idea for my situation?
2. Would it be best to start with the max of all 4 drives 4X2TB or add drives in pairs as i need them?
3. What Operating System would be best fit with all my devices/needs?
4. How extensive of a Power Supply should i have? I want something stable that can handle being powered on for long periods of time without costing an arm and a leg.
5. Since i will mainly be streaming media is there anything else i should consider? Ex. X amount of RAM/Video card?

I am unsure what path to take and looking for any help you wonderful people can offer me. I thank you for your insight and time.

 

[drescherjm]

1. What/if any RAID would be the best idea for my situation?

I would not use raid for an HTPC or even most home PC usage. For business or work that is a different story. In either case RAID does not in any way remove the need for a backup. RAID is about ensuring uptime during a disk failure.

2. Would it be best to start with the max of all 4 drives 4X2TB or add drives in pairs as i need them?

Without using raid I would focus on 5XXX rpm 4TB drives. These can be had for $150 or so on sale from time to time. And B&H has internal hitachi 4TBs for sale for $151.

3. What Operating System would be best fit with all my devices/needs?

For my linux based HTPC (mythtv) I use 64 bit gentoo linux for the last 8 or so years of nearly 100% uptime. Although I do not recommend this to those who do not know linux. For the storage I am using individual 2 and 4 TB green disks. Mythtv can handle any number of storage devices you give it. The software automatically balances your storage among the storage devices it has in the pool. I do now backup the videos to 4TB externals. Although for the first 9 years I did not include recordings in my backups.

4. How extensive of a Power Supply should i have? I want something stable that can handle being powered on for long periods of time without costing an arm and a leg.

I use a 650W antec that I purchased when I moved to a core2quad in 2008. That is plenty of power for the 6 to 8 internal hard drives, tuner cards, usb3 card, nvidia gpu...

5. Since i will mainly be streaming media is there anything else i should consider? Ex. X amount of RAM/Video card?

Streaming video is not that taxing even hd video. Although this may depend on what software you use for your htpc. I can easily stream multiple HD streams while recording multiple HD streams. As for the video card I purchased a nVidia fanless 520 to replace my fanless gforce 8400 that died a year or so ago. nVidia cards are recommended with mythtv because the mpeg hardware video acceleration is more supported over Intel or AMD based GPUs.

 

[staticlag]

Yeah your big mistake is thinking your data will be safe if you decide to RAID them.

RAID is about uptime and performance, not about data safety.

From what you are posting it sounds like cost is an issue. Sure anyone here could recommend a stellar system but it might cost $3-4K.

What type of price range are we looking at that you are willing to spend?

 

[Shifte]

Thanks for the reply(s)!

To clear up some confusion. I may build a HTPC in the future as my old one died but im living without it for now. I ONLY am looking to store and stream to multiple devices/PCs. When i think of HTPC i think of it as recording shows/watching movies/surfing the web from the couch/streaming content from a server.

Cost really isnt an issue. 3-4k is doable but would require me to wait a little longer than i am ready to. 1K easliy doable now, 2k in a few weeks etc. I just want to make sure that i have room to grow, and that it meets my needs.

Maybe i should just build a NAS? and not a server at all?

 

[Phog]

unRAID seems like a good fit for your situation. Not terribly technical, good value for your money, best utilization of space, and a stellar box for streaming video to HTPC's.

 

[Bob_Dole]

To jump in on the same topic. I'm getting an HP N36L to put 5x2tb disks in it, I was thinking of going raid5, but it's all downloaded media.

Would JBOD just be better for me, of is the Raid5 worth the 1 disk of space incase I lose a 2tb drive...

 

[teh_chem]

OP, it sounds like all you're asking for is a reliable backup of your files. RAID--despite whatever redundancy methods it employs--is not a backup solution. The main use for RAID is for hardware redundancy where some computer cannot afford any down-time (the data on the computer should be backed up elsewhere). Your best bet is to follow the rule of 3. 3 copies of your data. (1) the primarily-used drive(s). (2) a local backup, like to an external hard drive or NAS. And (3) an off-site copy, whether it's on an external hard drive you supply and store elsewhere, or whether it's via a cloud storage solution.

Don't bother with RAID on your computer unless you just want to play with the idea of RAID, and are willing to potentially lose data. A NAS is the better option--provided you have an other reliable backup of the data.

Bob_Dole--is your purpose to avoid down-time in case a hard drive fails? That would be the only real reason to use raid5. Also, JBOD is not entirely useful for most storage uses. And in Windows (7 and higher--perhaps Vista too), you can just span multiple physical disks and form one logical disk in the OS.

I don't mean this in an offensive way, but I think most people who decide on RAID for anything other than an enterprise-level service or something they just want to play around with are actually better suited with just a reliable back-up solution (again, RAID is not a reliable backup solution). The biggest issue (IMHO) with RAID is that it's dumb; it doesn't know if data it's mirroring or in parity is actually correct data. For example you have a power surge and it corrupts your primary hard drive. the RAID controller will just blindly mirror/copy the data, and now you have two messed up copies of your data. Happened to me (I learned the hard way...).

 

[Bob_Dole]

Bob_Dole--is your purpose to avoid down-time in case a hard drive fails? That would be the only real reason to use raid5. Also, JBOD is not entirely useful for most storage uses. And in Windows (7 and higher--perhaps Vista too), you can just span multiple physical disks and form one logical disk in the OS.

Good point, I'll probably be running linux on the N36L but I might use the LSI 8708 controller I have if it can just work in adding more drives to my system. Realistically, if I lose a drive, that would suck, but it'll be for tv/movies/ steam game storage, externally.

Thanks for the info about RAID, I appreciate it!

 

[kenned3]

OP, it sounds like all you're asking for is a reliable backup of your files. RAID--despite whatever redundancy methods it employs--is not a backup solution. The main use for RAID is for hardware redundancy where some computer cannot afford any down-time (the data on the computer should be backed up elsewhere). Your best bet is to follow the rule of 3. 3 copies of your data. (1) the primarily-used drive(s). (2) a local backup, like to an external hard drive or NAS. And (3) an off-site copy, whether it's on an external hard drive you supply and store elsewhere, or whether it's via a cloud storage solution.

Don't bother with RAID on your computer unless you just want to play with the idea of RAID, and are willing to potentially lose data. A NAS is the better option--provided you have an other reliable backup of the data.

Bob_Dole--is your purpose to avoid down-time in case a hard drive fails? That would be the only real reason to use raid5. Also, JBOD is not entirely useful for most storage uses. And in Windows (7 and higher--perhaps Vista too), you can just span multiple physical disks and form one logical disk in the OS.

I don't mean this in an offensive way, but I think most people who decide on RAID for anything other than an enterprise-level service or something they just want to play around with are actually better suited with just a reliable back-up solution (again, RAID is not a reliable backup solution). The biggest issue (IMHO) with RAID is that it's dumb; it doesn't know if data it's mirroring or in parity is actually correct data. For example you have a power surge and it corrupts your primary hard drive. the RAID controller will just blindly mirror/copy the data, and now you have two messed up copies of your data. Happened to me (I learned the hard way...).

--

I'm going to have to disagree.
You state he should not use RAID, then recommend what is potentially the WORST raid level available (Raid 0, stripping). If you span a drive over multiple disks and lose ONE disk you lose ALL the data. That's right, even though only one drive died you lost however many drives were in the span worth of data.

Next, you recommend the rule of 3. How painful... When you add/remove data from one disk you have to do the same thing more on all your backups? Do you automate this with something like snapshots?

You are correct, RIAD is not a backup, but it is not "enterprise-level service" or to "avoid downtime". Its a means to protect data from disk failure and sometimes to gain performance.

Real world experience (home use, not enterprise work stuff).
I put my first raid system together back in 2009 (7 x 2TB disks) using solaris ZFS. Since then I have replaced the system three times (exporting and importing the disk array to the new system each time) and had a drive start to fail (I replaced before it died with zpool replace). I've lost no data, never had any data really at risk, and have incredible read/write speeds well in excess of the individual drives (think >500MB/s on spinning drives, not SSD's).

This part confuses me:
--
the biggest issue (IMHO) with RAID is that it's dumb; it doesn't know if data it's mirroring or in parity is actually correct data. For example you have a power surge and it corrupts your primary hard drive. the RAID controller will just blindly mirror/copy the data, and now you have two messed up copies of your data.
--

First, what is "dumb" the OS, the raid controller card?
Next, you don't seem to understand the difference between mirror and parity and seem to use them interchangeably. If you had a mirror, you had nothing to do with parity.

Here are some definitions (thanks wikipedia):
Mirror - In data storage, disk mirroring or RAID1 is the replication of logical disk volumes onto separate physical hard disks in real time to ensure continuous availability. A mirrored volume is a complete logical representation of separate volume copies.

Parity - Many RAID levels employ an error protection scheme called "parity", a widely used method in information technology to provide fault tolerance in a given set of data.


This "problem" you mention with hardware raid controllers is precisely why many on this board use ZFS. ZFS uses checksums at a block level and "copy on write" to avoid these issues. So in effect, ZFS is well aware of what it is reading/writeing vs what it is suppose to be doing.


In summary:

If you want to have one large "drive" which has the aggregate space of all your disks (minus one) you probably want to go with raid5 (or higher). This way you get (NumDisks-1)*DiskSize in space (you lose a drive for parity). Further you can have one disk die and not lose anything.

If you want to deal with 5 drive letters and try to manage the space among them manually, as well as lose all the data one one drive should it die then go with JBOD.

If you want to create a major accident waiting to happen, just span them all. One disk dies you lose everything (NOTE, DO NOT DO THIS)...

As people here have mentioned and its worth repeating, RAID is not a replacement for backups....

 

[Bob_Dole]

In summary:

If you want to have one large "drive" which has the aggregate space of all your disks (minus one) you probably want to go with raid5 (or higher). This way you get (NumDisks-1)*DiskSize in space (you lose a drive for parity). Further you can have one disk die and not lose anything.

If you want to deal with 5 drive letters and try to manage the space among them manually, as well as lose all the data one one drive should it die then go with JBOD.

Would I have that huge of an issue with all my drives in JBOD (in my NAS, wouldn't it just show as 192.168.1.*** and the files of all of the drives), and would raid 5 up my performance any?

 

[kenned3]

Would I have that huge of an issue with all my drives in JBOD (in my NAS, wouldn't it just show as 192.168.1.*** and the files of all of the drives), and would raid 5 up my performance any?

Well, lets assume you are using windows and you create 5 drives, and share them.
They would show up something like this:

\\192.168.1.***\share1
\\192.168.1.***\share2
\\192.168.1.***\share3
\\192.168.1.***\share4
\\192.168.1.***\share5

There might be a way to try to aggregate these but i am not aware of such a mechanism.

Among issues with the propsoal this is one. The other is what do you do when "share3" is nearly full and "share5" is almost empty? Its difficult to try to manage your free/used space. (I have been there, done that...)

This is one of the issues "raid5" attempts to address (storage pools with redundancy should a disk inevitably fail). It allows you to easily aggregate all the storage onto a single "drive".

On my system all the storage is on a single "pool" but I share them out on a per-use share (documents, tv, etc).

So all the available free space is available to any of the "shares". Meaning if i record a new TV show it goes to the pool and the overall free space in the pool is reduced. If i had 5 drives i would need a more complicate scheme to manage the free/used space. This becomes frustrating when say "share2" stores TV shows and gets full.

Now you can :
- Move the older stuff to another share
- delete something
- reconfigure the recorder to use some other drive
- create jobs to manage this for you by moving files around via a sheduled task.
etc.

Raid5 is provides a huge performance increase for READS. My system typically reads around 100MB/s on a single drive, but when reading from the raid5 pool it typically reads > 500MB/s.

Hope this helps.

 

[Bob_Dole]

Well, lets assume you are using windows and you create 5 drives, and share them.
They would show up something like this:

\\192.168.1.***\share1
\\192.168.1.***\share2
\\192.168.1.***\share3
\\192.168.1.***\share4
\\192.168.1.***\share5

Raid5 is provides a huge performance increase for READS. My system typically reads around 100MB/s on a single drive, but when reading from the raid5 pool it typically reads > 500MB/s.

Hope this helps.

Yup!

After the initial upload I'll be doing a lot of reads, but little (massive file) writing. and if it changes anything I'll probably be running Ubuntu off of a flash drive or FreeNAS.

 

[kenned3]

Hi guys!
QUESTIONS:

1. What/if any RAID would be the best idea for my situation?
2. Would it be best to start with the max of all 4 drives 4X2TB or add drives in pairs as i need them?
3. What Operating System would be best fit with all my devices/needs?
4. How extensive of a Power Supply should i have? I want something stable that can handle being powered on for long periods of time without costing an arm and a leg.
5. Since i will mainly be streaming media is there anything else i should consider? Ex. X amount of RAM/Video card?

I am unsure what path to take and looking for any help you wonderful people can offer me. I thank you for your insight and time.

Let me address your questions as well.

1. RaidZ (zfs equivalent of raid5) would be a good starting point considering you have 2TB drives. It allows one drive to fail and still not lose data. should a drive fail you need to replace it quickly as if another dies you lose it all.

2) use them all. Using 4x2TB drives would give you ~6TB usable (remember you lose one drive for parity). If you are building it from scratch you might want to reconsider how many drives you start with. HINT: losing one drive in four is 25% but one drive in seven is a lot less in terms of percent storage lost to parity. There are tradeoffs to consider and you dont want to do something like 10 data drives and one parity....

3) This is really subjective. A lot of people seem to be recommend things like unraid, or freenas. I personally use Solaris 11.1. It has a higher learning curve but it also has a lot of advantages we can get into if you like. I've used Linux and its software RAID a long time ago. I found it a bit unstable and moved to solaris when i decided i wanted ZFS as it was rock solid and had a lot more features. One thing linux can do which ZFS cant is reshape an array (expand by a single drive). While this is very time consuming in linux it currently is not possible in ZFS. You expand pools in ZFS by adding new vdev's (think adding 7 new drives to an existing 7 drive system).

4) I think people tend to go overboard on the power supply. 650W should be probably be more then enough. My original server had 7x2TB data drives (raidz) and 2x650GB OS drives (mirrored) and ran on a 650W supply for around 18 months before i replaced it with something better faster. My current system can hold 48 drives in a single unit and has redundant 1,300W supplies (there are two 1300 watt supplies). Obviously this is very underutilized but gives me plenty of room to expand.

5) if its only gong to be a streaming server get a cheap video card. Most Server OS's are not graphics intense. Spend the money on drives and memory. Make sure you get a decent network card (or a motherbord with a decent onboard card).

PS - My system does the same job as you intend to do (stream content)....

 

[kenned3]

Yup!

After the initial upload I'll be doing a lot of reads, but little (massive file) writing. and if it changes anything I'll probably be running Ubuntu off of a flash drive or FreeNAS.

OMG, its almost like i'm building my own stuff again... You are on the same path i was years ago.

If you are looking at ubuntu, look at ZFS on Linux.

https://wiki.ubuntu.com/ZFS

Really easy to install under Ubuntu and works well (my desktop runs ubuntu 13.04 and has a few ZFS drives attached).

Just watch what version of pools you create as it could limit your options in the future. For example if you upgrade to feature flags and use them you might not be able to export/import the pool to solaris should you decide to go down that path.

ZFS loves ram, so make sure you give it plenty to work with...

 

[Bob_Dole]

OMG, its almost like i'm building my own stuff again... You are on the same path i was years ago.

You, Bob_Dole's first Raid

ZFS loves ram, so make sure you give it plenty to work with...

Yeah I was reading this, going to pair it with 1gb/TB of storage.

Also as far as Software goes,would building the raid in the bios controller on my LSI 8708EM2 be better (if that's how it works) ?

Thanks, for all the help so far, I really appreciate it!

 

[kenned3]

You, Bob_Dole's first Raid
Yeah I was reading this, going to pair it with 1gb/TB of storage.

Also as far as Software goes,would building the raid in the bios controller on my LSI 8708EM2 be better (if that's how it works) ?

Thanks, for all the help so far, I really appreciate it!

I'd probaly just JBOD the disks on the controller and let ZFS do the raid, something like
zpool create vol0 raidz /dev/sda /dev/sdb /dev/sdc /dev/sdd etc...

Once you have the pool created it will automount at boot without you editing the fstab. This is among the many advantages of the ZFS model.

From there create a bunch of ZFS file systems for orginaizational reasons like this:
zfs create vol0/tv
zfs create vol0/movies
zfs create vol0/whatever else you need

Then share these out via NFS or samba. Since they are just zfs file systems on a pool they all share the pool space so vol0/tv cant fill up leaving vol0/movies with a ton of free space, they all share the available space.

By not using the raid features of the card if you need to upgrade in the future the raid pool is not dependent on the controller card. You also remove the dependency to replace the card if it dies with the exact same card, and any card related issues.

My pool was originally on 5 of the SATA ports the motherboard came with and a PCI sata card for the other two drives. Upgrade number one was to replace this with a SAS card to clean up the cabling, improve performance and remove the crap SATA card.

since they were just sata drives attached to a sata controller zfs didnt care that the sata controllers changed. Just that it was able to access the drives and determine they belong to a pool.

 

[Bob_Dole]

I'd probaly just JBOD the disks on the controller and let ZFS do the raid, something like
zpool create vol0 raidz /dev/sda /dev/sdb /dev/sdc /dev/sdd etc...

since they were just sata drives attached to a sata controller zfs didnt care that the sata controllers changed. Just that it was able to access the drives and determine they belong to a pool.

Ahh I see, I might go through the card so I can manually enable write cache and maybe speed things up a bit (at least for writing).

But ZFS makes a lot of sense.

 

[kenned3]

Ahh I see, I might go through the card so I can manually enable write cache and maybe speed things up a bit (at least for writing).

But ZFS makes a lot of sense.

with ZFS you wont be disappointed, It can do some pretty amazing things.

I'm currently streaming a THD V show off my "spare" system (I power down the main system in the summer).

I've decided to run a scrub to make sure the data is safe. Its a single 3TB drive (not raidz).

Scan reports:
295G scanned out of 1.12T at 162M/s, 1h29m to go

I ran it again a few minutes later to show its not just a 'burst'

343G scanned out of 1.12T at 163M/s, 1h24m to go

This crazy-high read speed is while Streaming HD content with no stuttering or delays. How many other file systems can run a "check" while still online and maintain these speeds?

 

[Red Squirrel]

With the low reliability of hard drives these days, and the sheer amount of data they can hold, I cannot imagine not going raid or equivalent, such as ZFS. Raid, along with good backups is the only way I would do things for data storage. Without raid, you can have all the backups in the world, but you'll still lose anything since the last backup, and you will be dead in the water until you order a new drive, then copy all the data over from the backups, fix all the permissions so it works with everything, etc... the downtime is just not worth it, especially when you have better things to do than to recover a server. Raid with proper monitoring and backups is pretty much set and forget, and if a drive fails you pop a new one in and let it rebuild and you are golden, without any downtime.

 

[teh_chem]

--

I'm going to have to disagree.
You state he should not use RAID, then recommend what is potentially the WORST raid level available (Raid 0, stripping).

Please identify where I recommended RAID0.

If you span a drive over multiple disks and lose ONE disk you lose ALL the data. That's right, even though only one drive died you lost however many drives were in the span worth of data.

I was not addressing using a spanned disk as a method of backup or redundancy or integrity. Only to have one logical disk. And I only mentioned it in the context of Windows because it's a simple software-level thing to do (no need for a disk controller, no need for pre-OS configuration). The user was asking about JBOD on a raid controller. I mentioned spanned in windows because it's the same thing, and does not require extra hardware, and can be set up quite easily (assuming the person is using windows). Not touting it as a GOOD option that accommodates everything from redundancy to backup to failure.

Next, you recommend the rule of 3. How painful... When you add/remove data from one disk you have to do the same thing more on all your backups? Do you automate this with something like snapshots?

Personally at home I use Cobian backup. Option set to incremental backup. Or alternatively it can be differential, and you can choose how many backups to keep. It's completely automated. My off-site is done via CrashPlan.

You are correct, RIAD is not a backup, but it is not "enterprise-level service" or to "avoid downtime". Its a means to protect data from disk failure and sometimes to gain performance.

Read back at what I wrote. I'll specify: RAID, implemented as it was intended (not just simple RAID0 or RAID1) is more of an enterprise-level service than it is a consumer-level service. And it is absolutely implemented to avoid system downtime in the event of a disk failure. Sure, it's also to have as many working copies of data in a redundant fashion. My point was that it is not a backup mechanism.

This part confuses me:
--
the biggest issue (IMHO) with RAID is that it's dumb; it doesn't know if data it's mirroring or in parity is actually correct data. For example you have a power surge and it corrupts your primary hard drive. the RAID controller will just blindly mirror/copy the data, and now you have two messed up copies of your data.
--

First, what is "dumb" the OS, the raid controller card?
Next, you don't seem to understand the difference between mirror and parity and seem to use them interchangeably. If you had a mirror, you had nothing to do with parity.

I meant that even if you are mirroring, and your controller is enforcing parity on some RAID level, it still doesn't know what the data was supposed to be in the first place. Re-read my example. Corrupted source data will be copied by the controller regardless of whether you wanted it to (in a "dumb" fashion). If you have a RAID1 array, and your working drive data becomes corrupted, the controller will copy over the data, not knowing that it's corrupted. And you end up with 2 copies of corrupted data (and no backup, assuming the user is counting on the second disk in the RAID1 array to be their backup).

This "problem" you mention with hardware raid controllers is precisely why many on this board use ZFS. ZFS uses checksums at a block level and "copy on write" to avoid these issues. So in effect, ZFS is well aware of what it is reading/writeing vs what it is suppose to be doing.

ZFS doesn't work reliably on hardware RAID levels. Regardless, the assumption being made is that "something" knows that the original data is intact or corrupted; which cannot always be realized regardless of ZFS or any other mechanism. And I only mention this relative to the idea that some people think RAID is a backup mechanism; which it is not, and I can see that you agree with that point.

 

[kenned3]

Didn't you write this:

"you can just span multiple physical disks and form one logical disk in the OS."
What do you think "spanning multiple disks to form one logical disk" is if its not also known as raid0?

" I mentioned spanned in windows because it's the same thing,"
JBOD and Spanning are NOT the same thing and you failed to mention the serious downside of spanning disks. Its fine to discuss things like spanning but like all the raid levels it has a specific use case, and a specific set of risks.

"I meant that even if you are mirroring, and your controller is enforcing parity on some RAID level"
Mirroring doesn't have parity. It writes the same data on more then one disk. Nothing more complicated.

"ZFS doesn't work reliably on hardware RAID levels."

See my post when he asked about using his RAID controller and I recommended against it, and just let ZFS manage the disks.

 

[Bob_Dole]

With putting the controller in JBOD, can I still use it's 128mb of write cache?

Toying with slowly adding the drives over time (spread out the cost ) and then building the raid when I have them all... Then again, it would be silly to put data on those drives before throwing them into a RAID.

And even though I'll lose 1 drive to parity, I'd prefer it. the data isn't important, but I don't want to lose 2-4tb of data (Plus it'd be fun to mess with raid)

 

[drescherjm]

I'd prefer it. the data isn't important, but I don't want to lose 2-4tb of data (Plus it'd be fun to mess with raid)

What about the risk of loosing the entire array? Yes I know it is a small risk however even zfs users have lost their entire array. The only way to be protected is to make a backup (or two) and keep the backup(s) offline and in a safe place.

 

[Bob_Dole]

What about the risk of loosing the entire array? Yes I know it is a small risk however even zfs users have lost their entire array. The only way to be protected is to make a backup (or two) and keep the backup(s) offline and in a safe place.

This is a very valid point, so I guess it's weighing one option over the other.

I think raid 5/z1 is probably the best option as far as the ability to lose 1 disk and keep trucking until I get a replacement.

It'd be that or run them as single disks, and manually sorting things per disk (not a huge issue though, but I think raid would be neat).

Also I might add a small SSD to ZFS (if I do that versus hardware raid) for write cache, as opposed to the controllers onboard write cache.

 

[kenned3]

What about the risk of loosing the entire array? Yes I know it is a small risk however even zfs users have lost their entire array. The only way to be protected is to make a backup (or two) and keep the backup(s) offline and in a safe place.

Do you have any references to users losing ZFS array's? As a fan of ZFS i'd be interested in reading about why they lost data and what happened.

I've yet to come across actual reports related to data loss on ZFS pools (redundant or otherwise).

 

[kenned3]

With putting the controller in JBOD, can I still use it's 128mb of write cache?

Toying with slowly adding the drives over time (spread out the cost ) and then building the raid when I have them all... Then again, it would be silly to put data on those drives before throwing them into a RAID.

And even though I'll lose 1 drive to parity, I'd prefer it. the data isn't important, but I don't want to lose 2-4tb of data (Plus it'd be fun to mess with raid)

Probably not a wise idea. One of the things about ZFS is it wants to control the disk read/write caches. It needs to know when it has written to disk, it actually is done. ZFS uses the ARC cache which is either memory or can be SSD's for its caching mechanism. This is why ZFS wants memory....

One of my systems has 16GB ram, of this around 12 ends up going to the ARC (http://en.wikipedia.org/wiki/Adaptive_replacement_cache)

 

[Bob_Dole]

Probably not a wise idea. One of the things about ZFS is it wants to control the disk read/write caches. It needs to know when it has written to disk, it actually is done. ZFS uses the ARC cache which is either memory or can be SSD's for its caching mechanism. This is why ZFS wants memory....

One of my systems has 16GB ram, of this around 12 ends up going to the ARC (http://en.wikipedia.org/wiki/Adaptive_replacement_cache)

Well, I'll go ahead and test it as soon as I get the N36L and see where what stands where.

I'll go ahead and try ZFS and Hardware Raid and see how it does, probably going to run Ubuntu 12.04 or something along those lines unless there is a better option (Bob_Dole loves his GUIs)

 

[teh_chem]

Didn't you write this:

"you can just span multiple physical disks and form one logical disk in the OS."
What do you think "spanning multiple disks to form one logical disk" is if its not also known as raid0?

Spanning/JBOD. Not the same as striping (RAID0). Regardless, getting back to the point, I was still not mentioning this as any sort of backup, which has been the theme.

" I mentioned spanned in windows because it's the same thing,"
JBOD and Spanning are NOT the same thing and you failed to mention the serious downside of spanning disks. Its fine to discuss things like spanning but like all the raid levels it has a specific use case, and a specific set of risks.

Can you explain where JBOD differs from a spanned logical drive?

"I meant that even if you are mirroring, and your controller is enforcing parity on some RAID level"
Mirroring doesn't have parity. It writes the same data on more then one disk. Nothing more complicated.

Are you sure hybrid RAID doesn't have parity? Like 5+3? Again, point being, even with parity and mirroring, it still doesn't address data corruption at the source. Though you brought up ZFS.

"ZFS doesn't work reliably on hardware RAID levels."

See my post when he asked about using his RAID controller and I recommended against it, and just let ZFS manage the disks.

Point taken. Still not a backup method, so I'm losing the significance of this.

 

[brutalizer]

Yeah I was reading this [ZFS loves RAM], going to pair it with 1gb/TB of storage.

This is a misconception that is not really true. If you are going for ZFS deduplication, then you need at least 1GB RAM for every TB of disk storage. If you are not going to dedupe (which you should not do, because the ZFS dedupe engine is so-so right now) you dont need that much RAM. 4GB RAM is enough. If you have more RAM the very efficient disk cache will be larger. If you have 8GB RAM, often the disks will not be touched at all. If you have less RAM, the performance degrades to disk speed. Which is often ok for most people. I myself ran ZFS and Solaris on 1GB RAM PC for over a year without problems. It was slow because I used a 32 bit Pentium 4 cpu (ZFS is 128 bits and prefers 64 bit cpus) but in no way I had problems.

So, 1GB RAM for every TB disk - is only recommended when you use ZFS dedupe, which no one should use anyway. Instead, the Enterprise recommendation is to apply Compression on everything. ZFS compression is very cheap and often it can increase performance. It takes more time to read 150 blocks from disk, than it is to read 100 blocks and decompress to 150 block in RAM. Oracle recommends turning on compression on everything, because it is so stable and cheap. So use compression even if you are storing media files, because if ZFS can not compress the data (it might be media) the compression will notice it and early abort and no performance is lost.

4GB RAM would be fine for a server, if you can accept disk speed and no fancy disk caching. 8GB RAM would be fine too.

 

[bAMtan2]

if you have to ask this question, the deepest you should go with raid is RAID 1 with high end hardware raid, or MAYBE in software with the OS you use all the time.

if not that, then some of the easy file pooling and mirroring software for windows, whatever the heck they're called

 

[Bob_Dole]

...

 

[Justintoxicated]

Google "Drive Pool" and "Drive Bender". Will Redundancy at the folder level work for you? You can grow or shrink the virtual disk at any time, and you will get Raid-1 like redundancy at the folder level, (Meaning you can put "I want everything in this folder to span 2 disks", but this folder over here does not need redundancy ) maximizing space instead of wasting it. It's not a RAID solution which IMO is a good thing, much less complicated, you get one giant expandable disk, and redundancy. This does not use Parity, it actually duplicates the entire folder of data across multiple disks, this has both advantages and disadvantages.

I'm in the exact same boat right now and I am looking at MSE 2012 + drive bender or drive pool, I'm not sure how the backups work yet but if they are anything like WHS V1 then I will love it. WHS V1 keeps incremental backups of all Windows PCs on my network without wasting unnecessary space. I have rolled my PC's back to a previous date MANY times thanks for my GF getting viruses in her emails, data corruption, used to to change OS drives etc. TO backup you simply put the restore CD into the PC you want to restore, put your password to connect to the WHS box, and pick the date you want to restore from.

 

[tran1981]

Raid for redundancy. Still always have backups.