RAID Pointless in a Home Media Server?

[poofyhairguy]

Ok, before I start I know some of you will think I am crazy. I know as a computer geek my logic should be that RAID exists and for that reason alone I should use it.

But I have been looking to put together a new (probably Linux) media server and for the life of me I can't think of one good reason to use RAID.

Reasons I don't think RAID is worth it for me:

1. I am not buying my HDs all at once and I don't plan on just buying disks of all the same size for my server (I hope to go larger as large drives get cheaper), so it seems RAID 5 is useless to me.

2. I really don't care about redundancy. Not a single thing will go on my server that I can't just redownload/rerip, and the few things that would be a pain to redownload I am backing up on an external drive.

3. I don't like the idea of being at the mercy of software RAID (and I don't want to drop a few hundred bucks on a decent RAID card). I like the idea that if one of my drives (or my main OS drive) dies I can easily boot up a Linux Live CD and get at my data without having to rebuild a RAID and pray.

4. The speed advantages given by some forms of RAID are not needed. I recently tested (on my desktop) streaming 30GB+ Blu Ray rips to both my TV frontends at the same time using NFS (which I will use) and everything worked great. I don't see me ever having a use more demanding that that, so the speed advantages of some forms of RAID don't help me.

5. (The Big Reason) It seems like a huge point of RAID (or JBOD or WHS or Unraid or Drobos) is to make it so that all your different hard disks act as a single drive. But the media frontend software I use (XBMC) allows me to add in a ton of different directories as sources, and then it scrapes those directories into a single library. Having all my media seem like they are on one disk is useless to me.

So what do you all think? Am I missing something big here? Or are you just shocked that I am considering building a server without RAID and I need to turn in my geek card?

Advice is very welcome.

 

[RebateMonger]

Sounds good to me. Only you can decide what degree of complication and risk you are willing to take with your data storage.

I will point out that having to re-rip media can be a pretty time-consuming task, so there IS some value to having redundancy or backups. But there is a definite cost, both in dollars and in complexity, to add RAID or backups.

 

[Emulex]

i say screw raid. no need for it:

2TB (200gb C:\) 1.8TB C:\mp3
2TB c:\tv
2TB c:\movie
2TB C:\kids

etc...
mount points ftw !

 

[pjkenned]

2. I really don't care about redundancy. Not a single thing will go on my server that I can't just redownload/rerip, and the few things that would be a pain to redownload I am backing up on an external drive.

So let's say you have a small 2TB media collection, and rip from optical disks. At a minimum, assuming you only do straight 40GB optical disk rips (a popular size for blu-ray media) that's 45-46 disks swapped out (a 2TB drive is about 1.8TB usable). That is probably 2 mintues per disk of your time for swapping media and etc (most likely a bit more). Of course, I'm assuming the media is sitting next to the PC. Add, at minimum, 40min/ disk or 30 straight hours to get your media ready. That, of course, is after you spend a few days waiting for an RMA, or go buy another drive in the meantime.

So minimum cost for a non-redundant 2TB of media disk dying is:
1.5 hrs of your time
3-5 days minimum RMA turn time with advanced replacement
30 hours and lots of coffee after that you'll be up and running at 100%

If you instead are ripping music CD's, you then have many more disks than 45 I'm guessing.

Also, removable drives are of course, more prone to corruption/ breaking as you are transporting them + you need to unplug them.

Raid or WHS comes down to ease v. lower end solutions and the cost of downtime. Even restoring from backups takes a while. Remember, restoring 2TB of data at 100MB/s (which is pretty good average speed) is 5.5 hours of non-stop transfer.

When I was in college, quite a while ago, I had a hard drive that was backed up fail on me the night before a final project was due. I of course had the drive backed up, but it took 9 hours to restore the backups. 9 hours later, it was early morning and I realized that two new files that I had been working on the evening before were not backed up. It was a bad day.

Now if you have something like Windows Home Server and drive extender... then I could see not needing raid and that does do OK with different sized disks. If you are just streaming media, one of those small Ethernet/ USB "NAS" things may even work.

 

[poofyhairguy]

Thanks for the advice everyone. I will probably stick to individual drives plus external backup for the stuff I like the most. I really appreciate the feedback.

 

[Blain]

RAID 6 offers performance as well as redundancy.

 

[Emulex]

performance? let's compare write performance of raid-6 to a single jbod drive

 

[Mark R]

performance? let's compare write performance of raid-6 to a single jbod drive

When I tried it on a 4 drive array, it RAID-6 was noticeably faster than a single drive for large sustained transfers (e.g. media streaming) for read and write. Small file writes were single drive performance, possibly even worse though.

 

[poofyhairguy]

RAID 6 offers performance as well as redundancy.

And you have to buy at least four disks at once to get it going from what I understand.

I like the idea of having fast I/O speeds on my server, but aren't these large HDs fast enough to serve multiple clients?

Even a 5400 RPM Western Digital 2TB Green drive has sustained read speeds at or above 100mbs. That means that on that one drive I can serve two clients a copy of an Avatar-sized Blu Ray (which has 48 mb/s peeks).

I don't ever plan to have more than two clients ever going at once, and I plan to buy 7200 RPM drives so I have a little more headroom. I can understand needing more if I had kids or something (and therefore more clients) but if I get to that stage in life I will just build a new media server for that need (in the future when buying four 2TB drives at once costs way less!).

I really appreciate the comments, as I do feel a little lost sometimes. Every time I go to look up building a media server all the guides assume a RAID setup- as if you are a fool for wanting to do anything differently. I just want to make sure I am not a fool...

 

[RebateMonger]

I like the idea of having fast I/O speeds on my server, but aren't these large HDs fast enough to serve multiple clients?

On a networked server with only a few clients, read speeds across the network will be the same whether you use a single disk or a multi-disk striped array. Write speeds will be slower with the striped RAID array unless it's RAID 0 or RAID 10.

 

[poofyhairguy]

On a networked server with only a few clients, read speeds across the network will be the same whether you use a single disk or a multi-disk striped array. Write speeds will be slower with the striped RAID array unless it's RAID 0 or RAID 10.

Thank you very much for the information. I think I feel confident enough to start now!

 

[Blain]

performance? let's compare write performance of raid-6 to a single jbod drive

Let's not and just say we did.

Reading data is the main function of a media server.

 

[Mark R]

Even a 5400 RPM Western Digital 2TB Green drive has sustained read speeds at or above 100mbs. That means that on that one drive I can serve two clients a copy of an Avatar-sized Blu Ray (which has 48 mb/s peeks).

The drives can sustain 100 MB/s (800 Mb/s) - but one drive will barely sustain 2 high-bit rate blu-ray streams (up to 48 Mb/s). It will do it, but it'll struggle with a 3rd task (e.g. running a download/torrent, etc.)

The reason is that seeking on an HD is slow - when client 1 requests data, the drive has to seek to it, read it, then seek back for client 2. These are often relatively small reads (e.g. 256 kB). However, the amount of lost data transfer time (up to 25 ms on a 5400 rpm drive - 2 seeks) due to seek speeds, can be 10 times that (equivalent to have transferred 2-3 MB).

The result is that when serving 2 clients, the drive would only just manage to sustain 2 streams.

One advantage of RAID (as long as you have big chunk sizes - 512 kB or 1024 kB), is that the chunks (and therefore seeks) get spread out among the drives. Some reads go to drive 1, some to drive 2, etc. Therefore each drive spends less time seeking, and multi-streaming read performance is massively increased. One catch is that with small chunk sizes you may not get this boost - if a client requests 256 kB, and the chunk size is 32 kB - then the read will get broken into lots of small reads which will be distributed to all the disks - the result, all the disks seek simultaneously and all get lots of time wasted.

 

[Emulex]

i bet most cheap forms of raid would suck trying to stream two blu-ray while pulling 50 megs in on torrents and unrar'ing them lol.

 

[taltamir]

Ok, before I start I know some of you will think I am crazy. I know as a computer geek my logic should be that RAID exists and for that reason alone I should use it.

But I have been looking to put together a new (probably Linux) media server and for the life of me I can't think of one good reason to use RAID.

Reasons I don't think RAID is worth it for me:

1. I am not buying my HDs all at once and I don't plan on just buying disks of all the same size for my server (I hope to go larger as large drives get cheaper), so it seems RAID 5 is useless to me.

2. I really don't care about redundancy. Not a single thing will go on my server that I can't just redownload/rerip, and the few things that would be a pain to redownload I am backing up on an external drive.

3. I don't like the idea of being at the mercy of software RAID (and I don't want to drop a few hundred bucks on a decent RAID card). I like the idea that if one of my drives (or my main OS drive) dies I can easily boot up a Linux Live CD and get at my data without having to rebuild a RAID and pray.

4. The speed advantages given by some forms of RAID are not needed. I recently tested (on my desktop) streaming 30GB+ Blu Ray rips to both my TV frontends at the same time using NFS (which I will use) and everything worked great. I don't see me ever having a use more demanding that that, so the speed advantages of some forms of RAID don't help me.

5. (The Big Reason) It seems like a huge point of RAID (or JBOD or WHS or Unraid or Drobos) is to make it so that all your different hard disks act as a single drive. But the media frontend software I use (XBMC) allows me to add in a ton of different directories as sources, and then it scrapes those directories into a single library. Having all my media seem like they are on one disk is useless to me.

So what do you all think? Am I missing something big here? Or are you just shocked that I am considering building a server without RAID and I need to turn in my geek card?

Advice is very welcome.

I agree with all points. and actually you missed a few more reasons why RAID is bad...

however, in regards to point 2... How much is your time worth? measure how much of your time it takes to rip / download stuff... I use RAID1 (pure software... not the mobo crap, but the stuff that is 100% software from within the OS... i use ZFS)
Anyways, if you see how long it takes you to collect 1gb of "stuff" (ex: 2gb per hour), divide the total amount of stuff you have (ex: 2000gb / 2gb/hr = 1000hr) this give you how many hours it will take to recollect it all... now divide the cost of that much extra storage by the hours (aka, going from enough drives to store 2000gb to enough drives to store 2000gb in RAID1... aka, the price of one 2TB drive... currently 150$)
you get 150$/1000hr = 0.15$/hr
My time is worth more than 15 cents an hour thank you very much... so I use RAID1 on my "junk"... You can simply adjust the above figures and calculate it for yourself, but I would be surprised if you would get a very different result.

 

[Dark Shroud]

Just a note a new version of WHS is in the works, it's code named Vail.

http://www.winsupersite.com/server/whs_vail_storage.asp

 

[poofyhairguy]

however, in regards to point 2... How much is your time worth? measure how much of your time it takes to rip / download stuff... I use RAID1 (pure software... not the mobo crap, but the stuff that is 100% software from within the OS... i use ZFS)
Anyways, if you see how long it takes you to collect 1gb of "stuff" (ex: 2gb per hour), divide the total amount of stuff you have (ex: 2000gb / 2gb/hr = 1000hr) this give you how many hours it will take to recollect it all... now divide the cost of that much extra storage by the hours (aka, going from enough drives to store 2000gb to enough drives to store 2000gb in RAID1... aka, the price of one 2TB drive... currently 150$)
you get 150$/1000hr = 0.15$/hr
My time is worth more than 15 cents an hour thank you very much... so I use RAID1 on my "junk"... You can simply adjust the above figures and calculate it for yourself, but I would be surprised if you would get a very different result.

When I just did that same calculation on my old server, 1 to 1 redundancy would cost me about $2.5 an hour on it which seems reasonable considering I have no easy way to monetize my free time.

The two problems I have with such calculations are twofold:

1. With RAID there exists total risk because I can lose the RAID array. If one (RAID 5) or two (RAID 6) drives die I am in the clear, but if another beyond that dies I lose EVERYTHING. With independent disks the worst I can loose no matter how many disks die is only how much is on those disks. So the real calculation is not everything for independent disks, but worse reasonable case which is less than everything.

That makes the per hour charge much higher.

2. RAID treats all data as equally valuable. So my perfectly sorted Desperate Housewives directory (which CANNOT have downtime for household stability) that took me over 30 hours to rip perfectly is given the same priority as a pile of DVD rips that take the same space but I scripted as to only take maybe an hour or two of my actual time.

If the DVD rips go down I can take my time getting them back online, the DH goes down and I am up that night re-ripping disks. So it makes more sense to have a variable calculation depending on how much things are really worth, and for me I have backups for those things of very high value. (the DH directory is currently on a few drives).


Honestly the near perfect solution is something like Windows Home Server- lets me add different sized disks, lets me pick what get redundancy, lets me not worry about anything.

The only reason I am not using WHS now is the fact that my clients are mostly Linux (XBMC Live) and SAMBA performance on my Windows 7 desktop to my clients is too poor for me to use consistently.

If there existed something else that had those key features in Unixland then this thread never would have started, but the closest thing I have found is RAID which is lacking for all the above.

I am looking at ZFS now, as a RAIDZ allows for different size drives, but more and more it is looking like I might go with independent disks...

 

[pjkenned]

1. With RAID there exists total risk because I can lose the RAID array. If one (RAID 5) or two (RAID 6) drives die I am in the clear, but if another beyond that dies I lose EVERYTHING. With independent disks the worst I can loose no matter how many disks die is only how much is on those disks. So the real calculation is not everything for independent disks, but worse reasonable case which is less than everything.

If you lose three disks one a small-ish (sub 20TB) raid 6 array... before a hotspare rebuilds... that is pretty darn crazy. There's no reason you can't have multiple raid 6 arrays.

Also raid 6 writes with an IOP348 are not that bad you can actually get many 100's of MB/s pretty easily with cheap SATA disks.

 

[poofyhairguy]

Ok everyone, you win. I have been looking at benchmarks for software RAID6 and the performance seems completely acceptable to me.

I will wait to build a media server until I have 5 2TB drives accumulated, then I will build a software RAID Linux server. I like how I can just add disks after that and grow my array, and the parity hit at around 10 drives is pretty minimal. You are right that the concept of two drives dying is a stretch, and if I am honest with myself the parity will be welcomed.

Thanks for your help everyone...

 

[taltamir]

1. With RAID there exists total risk because I can lose the RAID array. If one (RAID 5) or two (RAID 6) drives die I am in the clear, but if another beyond that dies I lose EVERYTHING. With independent disks the worst I can loose no matter how many disks die is only how much is on those disks. So the real calculation is not everything for independent disks, but worse reasonable case which is less than everything.

well yes... I only do RAID1. I have done raid5, and even raid6... and I regretted it every time.

2. RAID treats all data as equally valuable. So my perfectly sorted Desperate Housewives directory (which CANNOT have downtime for household stability) that took me over 30 hours to rip perfectly is given the same priority as a pile of DVD rips that take the same space but I scripted as to only take maybe an hour or two of my actual time.

Well, you can make a cheap 2x750GB raid1 array for high priority stuff, and multiple non redundant disks for the low priority stuff .
I didn't say put EVERYTHING on raid...

Ok everyone, you win. I have been looking at benchmarks for software RAID6 and the performance seems completely acceptable to me.

I will wait to build a media server until I have 5 2TB drives accumulated, then I will build a software RAID Linux server. I like how I can just add disks after that and grow my array, and the parity hit at around 10 drives is pretty minimal. You are right that the concept of two drives dying is a stretch, and if I am honest with myself the parity will be welcomed.

Thanks for your help everyone...

I highly recommend against it.. also, unless you get a 300$ controller the performance on RAID6 (which I have used) will be SOOOO low compared to single disks that you will actually have problems with high def video... And copying stuff would be a PITA due to slowness...
RAID5/6 also have the write hole http://en.wikipedia.org/wiki/Standard_RAID_levels#RAID_5_disk_failure_rate
And RAID5/6 done via mobo means that the array will be lost whenever you reset CMOS or other bios events occur (upgraded bios? array lost, low battery? array lost... etc... granted it IS possible to recover it, you need to delete the array, then create a new array with exactly the same order of drives and stripe size and say NO when asked if you wish to zero the new array).

My recommendation to you is to have a bunch of lone drives... with one single RAID1 array for the most important stuff... so those desperate housewives perfect rips? they go on the raid1, most the other stuff goes on lone drives.

Or you can use a bunch of lone drives and simply use a PROGRAM such as secondcopy.com to automatically backup your important files from one drive to another... this is certainly more fine grained control on a per file basis.

 

[poofyhairguy]

Well, I was prepared to have to wait to RAID and you opened the can of beans again!

Seriously thanks for your comment though. I do want to say that I was planning on going with Linux software RAID, which I think is different than motherboard RAID.

More research was telling me though that to get decent performance in such a situation I would need a hardware RAID controller, which is one of the least appealing things about RAID.

Or you can use a bunch of lone drives and simply use a PROGRAM such as secondcopy.com to automatically backup your important files from one drive to another... this is certainly more fine grained control on a per file basis.

That is basically what my last file server was- just 9 x 1.5TB disks. I had a script to copy the stuff I wanted onto multiple disks (basically about 2.5 of those drives were for local backups). It served me well until it ran out, and the purpose of this thread was to discover if there was a better way before I started anew.

Man, this stuff is so tough. No wonder the unRaid and Drobo guys make a killing- the solutions for the problem of home media storage are either lacking, complicated or expensive! Your idea of having a RAID 1 partition is interesting though...

 

[taltamir]

Seriously thanks for your comment though. I do want to say that I was planning on going with Linux software RAID, which I think is different than motherboard RAID.

I know, that is true software RAID, OS based not mobo... mobo raid is not software raid at all, intel's implementation is more of a hybrid software hardware solution with the drawbacks of both and the benefits of neither.
I would not use linux OS based raid5 or 6 either... I would use linux OS raid1.
Only time you should even consider raid5 or 6 is if you have a 300$ controller

Man, this stuff is so tough. No wonder the unRaid and Drobo guys make a killing- the solutions for the problem of home media storage are either lacking, complicated or expensive! Your idea of having a RAID 1 partition is interesting though...

just to be clear, not a partition... Intel matrix raid allows you to actually raid a partition... I recommend against it. But yes, 1 set of raid1 for important stuff... make it drive D on that server. With drives E, F, G, H, I, J all being individual drives with no raid or anything...

That is basically what my last file server was- just 9 x 1.5TB disks. I had a script to copy the stuff I wanted onto multiple disks (basically about 2.5 of those drives were for local backups). It served me well until it ran out, and the purpose of this thread was to discover if there was a better way before I started anew.

I like raid1 better, but that is an excellent solution as well.

 

[Mark R]

I highly recommend against it.. also, unless you get a 300$ controller the performance on RAID6 (which I have used) will be SOOOO low compared to single disks that you will actually have problems with high def video... And copying stuff would be a PITA due to slowness...
RAID5/6 also have the write hole http://en.wikipedia.org/wiki/Standard_RAID_levels#RAID_5_disk_failure_rate
And RAID5/6 done via mobo means that the array will be lost whenever you reset CMOS or other bios events occur (upgraded bios? array lost, low battery? array lost... etc... granted it IS possible to recover it, you need to delete the array, then create a new array with exactly the same order of drives and stripe size and say NO when asked if you wish to zero the new array).

I don't think RAID 6 is that bad, provided that you have a decent software driver.

RAID 6 read performance is very good, especially for streaming. It's streaming performance is quite well approximate by (n-2) disks in RAID 0. In fact, the performance when handling multiple steams can be better than this - because the seeks are distributed among a greater number of disks. It's random writes that suck ass on RAID 5/6 - making them very poor choices for databases or transactional systems.

The write hole is an issue, but there are 2 workarounds:
1. 'Scrub' the RAID array after an unclean shutdown, and periodically (e.g. once a week). 'Scrubbing' is the name given to scanning the RAID to make sure all the parity is correct. Every disk is read, and every stripe is checked. Any incorrect parity is recalculated and rewritten. As the data hole is in the parity calculation, this should minimze the risks, by recalculating any parity that doesn't match the data.
2. Use a RAID driver and file system that use write barriers to guarantee ordering. Some filesystems use 'barriers' as a method of making sure that critical data is written in the correct order (to allow roll-back or journal recovery if interrupted by crash or power failure). A RAID driver that guarantees correct parity after a barrier, should mean that the write hole would only ever effect temporary data, and never valid file data or filesystem data. This feature is available on new linux versions using kernel 2.6.33 (this is still too new for some linux distributions, e.g. Ubuntu)

While a good quality hardware RAID card will do the things i've outlines above. Don't forget that hardware RAID cards aren't without a potential problem. If the card breaks, your data is inaccessible until you can get a compatible replacement (generally one manufacturer's card won't handle disks configured by another manufacturer's card). And even then, some replacement cards may not automatically detect the array, and may need to be manually configured with the exact same settings. Basically, make sure you know how to recover your data before you buy the card, and make sure that you can get spare cards if going this route.

One of the advantages of software RAID though, is that it can make use of the host CPU - which may be hugely faster than the CPU found on a hardware RAID card. Certainly with RAID 6, hardware RAID processors can start to become a bottleneck (by the time you get to 8-12 drives). Obviously, this isn't going to happen in a home server. But it is worth pointing out, that even top-of-the-range hardware RAID cards max out at about 800 MB/s write in RAID 6. By contrast, on a core i7 using linux software RAID 6, write speeds of over 1 GB/s are can easily be achieved using the same card with hardware RAID disabled.

 

[taltamir]

or you can just use a system that is not susceptible to said hole. like ZFS raid6 (raidz2)... but yes therea re certainly ways around it... that being said, I would not recommend it. RAID1 has been pure bliss, best solution. Backup via program? that works very nicely too...
raid5 and raidd6 I have regretted... I was warned against them but decided to try them anyways... both in hybrid mode (mobo controller) and full software (OS based)... I never tried those 300$+ pure hardware controllers because that price is ridiculous.

 

[bobross419]

You might consider looking in to Greyhole. It is still in late beta, but it is being released with Amahi servers now. It is a Linux version of WHS drive redundancy. I'm still waiting for full release before I build my server, but from everything I have seen it will do exactly what I want