Understanding RAID10 - help...!

[AT0MAC]

I am currently building a server, were I have a 2tb and a 3tb drive installed, next to a 64gb os ssd.
I want to build the data drives in RAID10, so I understand it's two RAID1's combined into one RAID0.

But do that mean I can start with one RAID1, later on expand with two more drives in a seperate RAID1 and then combine in RAID0 - without loosing data from the original RAID1 set???

The tricky part is what happens when I combine into RAID0, if it needs to clean all drives??

 

[giffut]

It ...

... seems by asking those questions you haven´t really laid out your strategy considering your server: What is its purpose. The RAID choice follows the purpose.

And just to be sure: RAID does NOT give you backup security, only delivers availability and throughput.

So: What is the purpose of your server?

 

[AT0MAC]

... seems by asking those questions you haven´t really laid out your strategy considering your server: What is its purpose. The RAID choice follows the purpose.

And just to be sure: RAID does NOT give you backup security, only delivers availability and throughput.

So: What is the purpose of your server?

The purpose of the server is to be used as a high-end home server. I will be streaming copies of my Blu-Ray in near-uncompressed quality to my mac computer so data throughput is important both when I import a new disc but also when I stream it out. Secondly it will function as my iTunes library but also my higher-than CD-quality FLAC library, so again throughput is important as a 24bit/192KHz file can easily take up 200MB.
Third, I don't like re-downloading so I keep copies of every software install that I use, sometimes really big files of a few GB.
Fourth, I have about 12.000 8MP pictures and a near endless pile of documents I can never ever re-make, so security is very important also.

All in all I want a combination of high bandwidth and high security, RAID1 gives me a copy of every file, and two RAID1's gives me two copies of every file - then extend that to RAID0 which doubles my bandwidth and I should be plenty home free with RAID10.

I want a box that contains all the above without having to rely on external drives. Beside all this the machine will also run Crashplan+ online backup of the most vital files, like pictures and documents.

So I ask again, If I wait till later and buy the last two drives and then make my RAID0 on top of the two RAID1's - will it delete the data off the original RAID1 set?

 

[Inconsequential]

No no no no no no NO.

RAID is NOT suitable for this. PERIOD.

I have my own NAS which does the following:

Disk 1: Timemachine
Disk 2: 2TB Photo Library Backup (150k RAW photos, 5-35MB each)
Disk 3-5: 6TB of Video Backups for streaming to PS3.

All protected by a parity disk using UnRAID.

Read and write speeds are of the order of the max Gigabit Ethernet can do (120MB/sec).

Using RAID for this is a silly idea.

 

[AT0MAC]

No no no no no no NO.

RAID is NOT suitable for this. PERIOD.

Using RAID for this is a silly idea.

And that is why?

RAID1 gives data backup and RAID0 gives speed, how can those two be a bad idea? As far as I can see, I can handle up to three drive failures before I actually start loosing data. Also, RAID1 doubles read speed and have normal write speed, so when adding a RAID0 on top of that I get quadruple read speeds and double write speeds, not bad...

...wait, it's only two drive failures right? Because the two separate RAID1s is part of the same RAID0 and therefor have half a file each - so a long as one drive in each RAID1 die then everything is ok, if two drives in the same RAID1 die i'm screwed. Correct?


First I actually wanted to use unRAID, but had trouble with getting my SuperMicro MB to boot from a USB stick so ended up selling it and redoing the whole thing.
Have looked into Linux, but never found my self at home, also Windows is a definite no no - so as I in my day to day work use Mac and know how they work inside and out I decided to use OS X Server with a few add on programs.

 

[slughead]

Okay so it's early in the morning so forgive me if I don't understand.

To have a RAID 10, you need a minimum of 4 of the same drive.

Basically, you have 2 RAID 0's setup and linked to each other through a RAID 1. This is very easy to do through diskutillity and I would STRONGLY recommend using that instead of a 3rd party controller card for reasons I could discuss if you want. Therefore if you have any 1 drive fail, you can continue to use your computer and wait for a replacement to arrive in the mail, at which point you can swap the bad drive and "rebuild." If you have 2 drives fail and they make up both drives of a striped pair, then it should be okay as well. However, if you have 1 drive die from each of the striped pairs, you've lost everything.

So you see, RAID 10 has great redundancy and unlike a RAID 5, if a drive fails it is still perfectly usable for all your normal tasks while you wait for a replacement.

I don't think it's possible to start with a RAID 0 and go to a RAID 10. It is DEFINITELY not possible to go from a RAID 1 to a RAID 10.

You should purchase all your drives for the array before you create it. And by the way, they should all be the same size, so you need to pick between 2 and 3 tb drives. Remember that since everything is totally backed up via the RAID 1, your capacity is cut in half. So 4 x 2tb drives in a RAID 10 is only 4TB.

You say you already have a SSD, you can put that in the optical bay. I use my SSD for holding my apps and booting. I don't even back up the contents--I'll just reinstall if it breaks.

 

[AT0MAC]

To have a RAID 10, you need a minimum of 4 of the same drive.

Basically, you have 2 RAID 0's setup and linked to each other through a RAID 1. Therefore if you have any 1 drive fail, you can continue to use your computer and wait for a replacement to arrive in the mail, at which point you can swap the bad drive and "rebuild." If you have 2 drives fail and they make up both drives of a striped pair, then it should be okay as well. However, if you have 1 drive die from each of the striped pairs, you've lost everything.

So you see, RAID 10 has great redundancy and unlike a RAID 5, if a drive fails it is still perfectly usable for all your normal tasks while you wait for a replacement.

I don't think it's possible to start with a RAID 0 and go to a RAID 10. It is DEFINITELY not possible to go from a RAID 1 to a RAID 10.

You should purchase all your drives for the array before you create it. And by the way, they should all be the same size, so you need to pick between 2 and 3 tb drives.

You say you already have a SSD, you can put that in the optical bay.

The chassis is a Fractal Design Node 304, so no optical bays - but there is space for 6 drives no matter what, so no worries (I can only squeeze in 5 because I have maxed out the case with other things).

Never mind...

I read somewhere that a RAID0+1 is not as reliable as a RAID10 (depending on if its two RAID0s in a 1 or the other way around).

RAID1 is the double vision version of HDDs:

Two of them gives me one copy of everything, but four times the read speed.
Then add a RAID0 layer and I get double the write speed also.

RAID0 is the speed option:

Two of them gives me zero copies but four times read and four times write.
Add a RAID1 and I get one copy of everything, but should give me only four reads and double writes as it will copy to the other side when the first pair of RAIDs is written to.


What option should I take as they seem to give the same benefits, RAID10 or RAID0+1??
EDIT: Just answered that one my self, no difference in a 4disc environment, but if you have the option always choose 10.


Back to the question, is RAID10 a good idea for a server?


.

 

[slughead]

I read somewhere that a RAID0+1 is not as reliable as a RAID10 (depending on if its two RAID0s in a 1 or the other way around).

Well that all depends. For a 4 drives setup, RAID 10 = RAID 01. For 6 drives, having 3 striped pairs of RAID 0 (10) has less capacity and speed than two striped groups of three (01) BUT less redundancy (less safety).

http://www.thegeekstuff.com/2011/10/raid10-vs-raid01/

Put simply: http://static.thegeekstuff.com/wp-content/uploads/2011/10/raid10-6disks.png has greater redundancy but less speed and capacity than http://static.thegeekstuff.com/wp-content/uploads/2011/10/raid01-6disks.png

For instance, let's say you're using 2TB Drives:

RAID01: With two groups of 3 striped drives put into a RAID 1. You will have the 6TB of capacity and the combined speed of 3 drives. You can have ANY 1 drive fail and the array will be fine, and you can have up to 3 drives fail provided they are all in a single striped set. It is possible for the array to fail if only 2 drives fail, provided one is from each set. This has less redundancy than the RAID 10, but a lot more speed and capacity.

RAID10: With THREE groups of 2 striped (RAID 0) drives put into a RAID 1. you will have TRIPLE redundancy but only 4TB and only the speed of a pair of drives in a RAID 0. You can have ANY 2 drives fail and the Array will survive. Also, you can have 4 drives fail, so long as the 2 remaining drives are striped together. It is possible to wipe out the array if one drive from each striped pair fails, but unlikely.

 

[AT0MAC]

Well that all depends. For a 4 drives setup, RAID 10 = RAID 01. For 6 drives, having 3 striped pairs of RAID 0 (10) has less capacity and speed than two striped groups of three (01) BUT less redundancy (less safety).

Great, as I am only using four drives I will use RAID10.

Is RAID10 a good idea for a server or do you say NO nononononono like Concorde rules did above without telling me why?

 

[Spacedust]

Great, as I am only using four drives I will use RAID10.

Is RAID10 a good idea for a server or do you say NO nononononono like Concorde rules did above without telling me why?

RAID10 is a very good idea for a server if you have exactly 4 drives, but you loose capacity of 2 drives.

 

[AT0MAC]

RAID10 is a very good idea for a server if you have exactly 4 drives, but you loose capacity of 2 drives.

I know... Actually when I think about it - would two RAID1 volumes have been a better option? I still loose two drives but I can save a little in power not spinning up all 4 drives all the time.

 

[DanielCoffey]

I think what is confusing some folks is that a lot of your data is archived for security and doesn't need the speed of access... the scanned documents for example.

Would you perhaps get a better result if you segregated your data into "stuff I don't need often but want to be secure" and "stuff I need really fast"?

Will an online backup plan be appropriate for this amount of data? Have you costed it and worked out how long it would take to get your data online?

 

[Inconsequential]

And that is why?

RAID1 gives data backup and RAID0 gives speed, how can those two be a bad idea? As far as I can see, I can handle up to three drive failures before I actually start loosing data. Also, RAID1 doubles read speed and have normal write speed, so when adding a RAID0 on top of that I get quadruple read speeds and double write speeds, not bad...

...wait, it's only two drive failures right? Because the two separate RAID1s is part of the same RAID0 and therefor have half a file each - so a long as one drive in each RAID1 die then everything is ok, if two drives in the same RAID1 die i'm screwed. Correct?


First I actually wanted to use unRAID, but had trouble with getting my SuperMicro MB to boot from a USB stick so ended up selling it and redoing the whole thing.
Have looked into Linux, but never found my self at home, also Windows is a definite no no - so as I in my day to day work use Mac and know how they work inside and out I decided to use OS X Server with a few add on programs.

If you have 4 drives and you loose 3 you loose everything. Infact, you can loose 2/4 drives and loose everything.

Proper servers may use RAID 10 because they need the performance whilst maintaining a good level of redundancy, you at home do not. Plus you are still limited to 1Gbps if you are using it as a NAS, which any decent HD manages over 50% of the capacity anyway.

Try UnRAID again, I've been using it for over a year and I have had no problems with it what so ever. WoL, Sleep and parity all works perfectly.

Also, if data backup is THAT important to you, then you should have multiple copies combined with offsite backups using different backup software.

One fire, flood or whatever and all your data on that supposedly good RAID 10 array is gone.

 

[deconstruct60]

...
Image
RAID10: With THREE groups of 2 striped (RAID 0) drives put into a RAID 1. you will have TRIPLE redundancy but only 4TB and only the speed of a pair of drives in a RAID 0. You can have ANY 2 drives fail and the Array will survive. Also, you can have 4 drives fail, so long as the 2 remaining drives are striped together. It is possible to wipe out the array if one drive from each striped pair fails, but unlikely.

No. "triple" redundancy would be three A's ( or B's , C's , etc.) in the diagram. There is still 2 in 0-1 or 1-0 .

If each 'group' is an independent SAS (or SATA ) channel then here is differences in redundancies along that aspect. In the previous 0-1 example even if the link to 'group 1' goes down you still have full access to all data. In the 1-0 example, you don't. Conversely, if read streaming lots of sequential blocks then 1-0 allows multiple concurrent reads in flight at once with very simple "read ahead" heuristics.

 

[NoNameBrand]

RAID1 gives data backup

No, no it doesn't. It gives availability. It means your server keeps on trucking while you go out and buy a replacement disk and slap it in.

Backup is having an extra copy on an extra volume for when something happens to a file on the first volume.

Both are important to people, but for most home users the latter is far more important.

You want to watch a video and you find out your hard drive died? With RAID, you don't do anything (in fact you might not know your disk died). With backup: Just point the sharing service at the backup and off you go. Five minute delay to your movie starting.

What happens if a video file gets corrupted or deleted: with RAID you have a corrupted video mirrored to two volumes. Yippie. With backup, restore from backup and you're good to go (if you caught the corruption in time).

If you really do want RAID, many RAID1 implementations will stripe reads since the data is on two disks, giving RAID0 read speed. I'd just go with that.

As for RAID10, yes you can build it in software on the Mac, yes you will lose all data on the drives involved. Build the two RAID0 sets, then build a RAID1 set from the resultant striped volumes. Just stop thinking you have backup with this setup.

 

[deconstruct60]

The purpose of the server is to be used as a high-end home server. I will be streaming copies of my Blu-Ray in near-uncompressed quality to my mac computer so data throughput is important both when I import a new disc but also when I stream it out. Secondly it will function as my iTunes library but also my higher-than CD-quality FLAC library, so again throughput is important as a 24bit/192KHz file can easily take up 200MB.

If there is just one Mac on the other side a fair amount of this doesn't matter since it would be a either/or situation. The files may be big but it is the speed at which they traverse the network to the Mac that is an issue. If the Mac is on WiFi that's the bottleneck. Even if on a wired 1GbE switch it is still an issue if a single destination.

Three RAID-1's would work: video , audio, misc . If you split the workload across multiple sets you can still get throughput even if one machine is leveraging video, while another is accessing audio, and another is doing a backup. You'll still have 99+% uptime so still have very high availablity (presuming the rest of the system doesn't have problems ) .

Putting everything into one big pile (single volume) only servers to drag in striping (RAID 0 , or striping with parity).

Video, even HD, isn't a big issue with proper buffering on either side of the network connection.

All in all I want a combination of high bandwidth and high security,

Bandwidth off the disk to the mITX board isn't going to matter much if it gets choked off by the network between this box and the Mac.

Similarly, if it is primarily just sequential block access the bandwidth even off a single disk is sufficient given the likely network cap. It would only be contentious access 2-3 computers conncurrently vying for different parts of the video library or a mixed workload on a single disk that would motivate a HDD bandwidth problem.

 

[slughead]

No. "triple" redundancy would be three A's ( or B's , C's , etc.) in the diagram. There is still 2 in 0-1 or 1-0 .

If each 'group' is an independent SAS (or SATA ) channel then here is differences in redundancies along that aspect. In the previous 0-1 example even if the link to 'group 1' goes down you still have full access to all data. In the 1-0 example, you don't. Conversely, if read streaming lots of sequential blocks then 1-0 allows multiple concurrent reads in flight at once with very simple "read ahead" heuristics.

Oh you're right, I totally effed that up

 

[AT0MAC]

I have now understood the point. Raid10 is not what I want.

Actually the original reason for building a server is that I have had a Buffalo NAS, a Linksys NAS and latest a Verbatim - all have failed and broke down.
I want a reliable way of keeping my data so backup is the very most important!


Therefor, I have thought about this type of solution:

I RAID1 the two 2TB main drives, so even if one drive fail I still have a functional copy, also I get double the read speed because everything can be read via both drives simultaneously.

RAID1 is very reliable, but I can still have corruption’s that make the two drives out of sync and in need of a rebuild and I could potential loose data - so I will setup my third 2TB drive to make a full clone of the 2TB RAID1 system every once a day.

Now I have secured any drive errors and software errors, but what about single file corruptions, still a possible thread. So my 3TB drive will be used as TimeMachine both for the RAID but also for my SSD were the OS will be housed.

Speaking about the OS, I have ordered a 64GB USB3.0 Integral Xcel flash drive to have an external boot clone copy of my OS made once a day, just like on the internal drives.



What do you guys think, possible, reliable and a good solution?

 

[NoNameBrand]

What do you guys think, possible, reliable and a good solution?

Much, much better.

 

[AT0MAC]

Much, much better.

Sorry to say that I can't exactly make what I wanted, due to very limited space in my chassis - so was only able to fit in 2x2TB+1x3TB+64GB.

As in, one primary data drive of 2TB, one daily clone of that drive also of 2TB, hourly 3TB TimeMachine partitioned into two so it also covers the 64GB OS SSD. The OS SSD is then cloned once a day to an external 64GB USB3 stick.

Not really as good, but very damn close.

Covers data corruption due to the time machine, covers drive failure due to the clones and also already have a brand new 2TB spare drive in original box laying on it's shelf here at home for the event one of the 2TB drives gives up.

Not bad not bad...

But no RAID...


Anyway, thank you a bunch for ideas and input to my original question, you guys are awesome!

 

[MatthewAMEL]

It was stated above, but RAID isn't backup. It's availability.

No matter what strategy/product you decide to go with, use the 3-2-1 rule.

3 copies of your data
2 on-site (production + local backup)
1 off-site (rotated media/crashplan, etc.)

 

[AT0MAC]

It was stated above, but RAID isn't backup. It's availability.

No matter what strategy/product you decide to go with, use the 3-2-1 rule.

3 copies of your data
2 on-site (production + local backup)
1 off-site (rotated media/crashplan, etc.)

Great advice!
And that is exactly the solution I have ended up with - and yes, I also realized raid is not the same as backup.