If RAID0 is so bad, what do you suggest?

[WooDaddy]

So,

I've been reading on many forums and it seems to me that many people recommend against using RAID-0. Those that are 'ok' with it suggest software RAID-0 instead of HW based. Other say use ZFS... which I can't use. So, what are my alternatives if I want must faster disk access (and don't want to buy an SSD... yet).

Here's my setups:

Main development PC:
- 2 WD 7200rpm 500GB HDs
- Core i7-920 on X58 mATX mobo with on-board Raid (ICH10R) (http://www.msi.com/index.php?func=proddesc&maincat_no=1&cat2_no=&cat3_no=&prod_no=1796).
- Running Win 7 64-bit as host OS - VirtualBox (for now) and running many other guest OSes. Note: not for server use, so I start up and tear down the guest OSes often

Main storage/backend server (yet to be built):
- 2 1TB 7200rpm HDs (mismatched hitachi and seagate... I know, I know)
- 1 80GB 5400rpm Host OS HD - maybe upgrade to a SSD later
- Atom D510 on a SuperMicro server ITX board with ICH9R northbridge.
- Not sure of the host os just yet. Maybe Nexenta, Openfiler or Xen with both running and a bunch other guest OSes for always on server use.
Here's the server I'm going to buy (http://www.newegg.com/Product/ComboB...t=Combo.489864)
* This may have the 2 1TB running in RAID-1 or XFS.


Maybe I'm stubborn, but I'm thinking about doing RAID-0 anyway on the main PC since ZFS isn't an option for Win7. All critical backups will be done to both the backend server and the cloud.

So to re-iterate the question(s):
1) Is RAID-0 REALLY that bad or do I need to be aware of the risks and have a good backup plan?

2) Is RAID-0 worse than a single drive solution? (It doesn't sound like it).

And.. for my non-Windows machine,

3) Is XFS the way to go or is RAID-1 ok considering I'm using a popular ICH9R raid controller.

 

[RebateMonger]

1) Is RAID-0 REALLY that bad or do I need to be aware of the risks and have a good backup plan?

2) Is RAID-0 worse than a single drive solution? (It doesn't sound like it).

Be aware of the risks and have backups.

A RAID 0 array will have two-or-more times as much risk of failing as a single-disk implementation. And, compared to a single disk, data recovery is much more complex.

 

[FishAk]

There is nothing wrong with running RAID 0, or any other kind of RAID, provided you have current backups including a valid image of your OS partition. You should have this anyway, as all disks are prone to fail at some point, and anything not backed up at the time of failure is at risk.

It's small consolation to know that you were running in single disk mode, and not RAID 0, when you loose that single disk.

 

[rmd2001]

RAID0 is not bad. It's actually very good - everything is faster. As RebateMonger said before chance of loosing data at least doubles.

But if you want reliability I would go with RAID10 - it's fast and chance of loosing data is very slim. In your case pick up 2 more 500GB drives (they are very cheap) and set all 4 as RAID10.

 

[ViviTheMage]

Nothing with raid, or raid 0 ... just keep backups.

also, I would NOT run a XEN node on that Atom server ... it'll dog like hell.

 

[Ka0t1x]

Add another 500GB, 3 Drive RAID 5.

You should get a mix of your redundancy (single drive) with some speed ontop of it.


Even with an SSD I still image my install (1:1), regularly.

 

[sub.mesa]

So,

I've been reading on many forums and it seems to me that many people recommend against using RAID-0. Those that are 'ok' with it suggest software RAID-0 instead of HW based. Other say use ZFS... which I can't use. So, what are my alternatives if I want must faster disk access (and don't want to buy an SSD... yet).

Your goal is faster I/O performance? Then a NAS might not be your best solution and a DAS would be more obvious.

NAS = Network Attached Storage (filesystem is maintained by the server; data travels over the network)
SAN = Storage Area Network (filesystem is maintained by the client; data travels over the network)
DAS = Direct Attached Storage (single system only; data travels via USB/eSATA/SATA/SAS/Lightpeak directly to workstation)

Think about what is most logical to you. a NAS would likely be limited to gigabit ethernet; 100MB/s is the maximum you can do so one drive could already mean you are network bottlenecked; and that is even assuming you get proper ethernet speeds which may not be the case for everyone.

A home-NAS is most used to store large files; mass storage. That means mostly sequential single-stream access. Moving to a NAS won't really help you with those; unless you invest in multi-gigabit networking (link aggregation, 10GBaseT, Infiniband, fiber). Where a NAS does help you with:
- data centralization (all your data under one roof tidely organised and accessible to all your LAN computers)
- free space aggregation (share all the free space of your disks under one unit; so you won't ever have to move stuff from A to B just to free up space.
- data security (protecting your valuable data with redundancy, multi copies, snapshots, checksums and/or encryption, to enhance its data security)
- automating backups (having two NAS boxes allows you to assign one as automated backup; for example making a snapshot and syncing the data from the main NAS to act as automated nightly backup solution)

Perhaps you want to use RAID0 in your workstation, for the speed. But backup the data on the NAS for data security. In that case you get the best of both worlds. RAID0 is not unsafe; it depends on how you utilize it. Your storage setup should always allow any particular disks to completely fail, without you running into any kind of trouble or major hassle that is prone to mistakes. If you can confirm this rule of thumb, then RAID0 may be a good solution for you.

Most people already use technologies similar to RAID0
It's kind of funny that some argue heavily against RAID0, but use their SSD without a backup. Current SSDs are essentially RAID0-ed 10 times (Intel), 8 times (Micron/Sandforce) or 4 times (Indilinx). They might also have dual channel memory, DDR memory, may run a SLI configuration and definitely have multi-processor technology. All those technologies fundamentally are 'interleaving'. This happens with your memory multiple times (the DDR, the dual channel, the bank interleaving; three 'RAID0' techniques all apply to your RAM. Of course this is implemented wildly different, but interestingly the performance characteristics of all this mechanics are surprisingly similar.

So don't dislike RAID0; it's great. Pure and simple.
Forget hardware RAID by the way, for the most part that's something of the past. Especially if you look beyond Windows, at least on your server.

If you want more gigabit performance, consider link aggregation. Cheap method to make a 4 gigabit link out of 4 realtek NICs or 3 realtek and 1 onboard gigabit. On Windows you need proprietary drivers for this; the Intel dualport server adapters for example. They aren't cheap though; your server should be able to use any network interface for aggregation if you opt for Linux/BSD OS.

For those unfamiliar, link aggregation ("lagg") can be used to increase performance of your network; sort of like RAID0 for your gigabit network; with 2 ports giving you theoretical maximum of 2Gbps and 4 ports = 4Gbps. Of course the scaling won't be 100% but still this is a cheap and relatively easy method to get higher networking throughput.

- Not sure of the host os just yet. Maybe Nexenta, Openfiler or Xen with both running and a bunch other guest OSes for always on server use.
Here's the server I'm going to buy (http://www.newegg.com/Product/ComboB...t=Combo.489864)

Test a few of these; if only in Virtualbox! The OS is very important and generally i recommend you decide this first before buying any hardware!

Maybe I'm stubborn, but I'm thinking about doing RAID-0 anyway on the main PC since ZFS isn't an option for Win7. All critical backups will be done to both the backend server and the cloud.

If you want the security of ZFS; then why don't you deploy it on your server? Then the server runs ZFS, and Windows 7 just puts files on there, via SMB/CIFS aka windows filesharing. All NAS-protocols control the filesystem on the server; the client doesn't really know or control the filesystem. So Windows 7 could be writing files on XFS, ZFS, JFS, ReiserFS, Ext3fs, etc; it won't care or know about it.

If you want ZFS you may need alot of memory; 4GB minimum and 8GB if you want higher performance (more complex setup). The Atom board is limited to 4GB though.

Don't use RAID1 without backup; you can achieve much higher reliability by using one of the disks as backup instead!

 

[WooDaddy]

Wow, thanks guys! I am always amazed at the helpfulness of the Anandtech community.

So, I think I'll still push forward with the RAID-0 on my main PC. I'm assuming I can move to RAID 5 in the near future as Ka0t1x suggested without have to start from scratch, right? Also, yes, I'll do SW RAID instead.

Vivi, good point. Actually, I've been dying to run Xen but I'll probably just play around with it on my main pc via VirtualBox until I get more familiar with it.

sub.mesa: WOW Thanks for all the info. I forgot ZFS needed all that RAM for it to be any good. I'll be starting off with 2GB at first... hopefully it won't be too bad. As for lagg, the Supermicro Atom mobo has dual gigE phys on it. Hopefully, it's supported. If so, then serving my VM disk images may be ok. Otherwise, I'll use it for nightly backups.

So instead of RAID 1, rsync or some other basic backup or cloning to another local drive is just as good?

Basically, what it sounds like is that RAID-5 for my PC is the way to go and either ZFS if I can afford it or RAID-1 + backup for the backend.
... or I could cheat and get a Drobo... (yuck)

Thanks a lot folks! If interested, I'll report back with my final setup. It'll be awhile though (1-2 weeks).

EDIT:
BTW I should mention, I'm really focusing on an Atom mainly for low power consumption. It's replacing an aging G4 Mac Mini running Fedora that is slow as molasses. I'm trying to keep my overall consumption down since my main is about 200W and the G4 is about 30-40W. My huge current draw is actually from my client's hardware and other heating costs, etc. Heck I'd actually like to go for the Athlon II Neo K3xx series since they seem to be really powerful compared to the D510 and D525 Atom's but I didn't find any server mobos with that on it.

 

[sub.mesa]

sub.mesa: WOW Thanks for all the info. I forgot ZFS needed all that RAM for it to be any good. I'll be starting off with 2GB at first... hopefully it won't be too bad. As for lagg, the Supermicro Atom mobo has dual gigE phys on it. Hopefully, it's supported. If so, then serving my VM disk images may be ok. Otherwise, I'll use it for nightly backups.

If you start at 2GB; make sure you pick a single 2GB memory module, so you can actually upgrade to 4GB. I think you would want to.

If the server is going to run Linux/BSD then you can use any NIC for lagg and also the two Intel NICs on that SuperMicro Atom board. But those NICs do not allow lagg under Windows i think; you need at least one Intel server adapter to be able to use the Intel lagg drivers for Windows. If you do, i'm told it also can serve a few onboard or desktop-edition NICs.

So getting lagg to work under Windows might be the hardest part, or at least most expensive.

So instead of RAID 1, rsync or some other basic backup or cloning to another local drive is just as good?

Yes, especially if you use snapshots. Then you make incremental backups.

That works pretty much like:

1. Sync from the main server
2. create snapshot 20101015
3. next day it syncs again
4. create snapshot 20101016

Now if the last sync turned out to be bad; all your files on the NAS were corrupt; and it copied and overwrite all your files with corrupt versions; then your data is still there! The power of snapshots:

zfs rollback pool/mydata@20101015

And all your data is back again; just like it was on 20101015. Snapshots are very powerful when combining them with backups. Intelligent use of redundancy, backups and snapshots can mean you never have to lose data or have data corrupted ever again!

Best if you setup this to work all automated; with email notification. That way it's maintenance free and you just check periodically if it still works. Until you get an email with error message that the pool failed; then you know you can jump into action.

BTW I should mention, I'm really focusing on an Atom mainly for low power consumption.

To be honest, you might just as well go for one of the low-power Core i3/i5 boards; the idle power consumption would be just as low except you have much more powerful CPU for when you really need it. That also means the CPU can idle quickly again because it completed its task so quickly; so today's most modern CPUs are VERY low-power in NAS servers.

For example, check this link:
http://www.servethehome.com/intel-core-i5-650-v-atom-n330-nvida-ion-review/

It lists the power consumption of the core i5 system under that of Atom solution. To be fair, that atom board may not be the most low-power of all Atom boards; but still you see that low-power doesn't mean you HAVE to sacrifice performance and go for atom. A good low-power CPU will do. That CPU can have 25W TDP, 45W TDP, 65W TDP, 95W TDP; that all doesn't really tell you how much it will be using when idling.

Modern CPUs have alot of methods of reducing power consumption when idling; the 32nm CPUs from Intel are great; and AMD doesn't do bad either with some very affordable solutions. The big bonus here is that you get ECC for free. Though you don't really absolutely need low-TDP, the Athlon 260u is a great low-power dualcore at 25W TDP max. Great for PicoPSU to keep maximum current low so you can use it to spinup HDDs when powering on the system (where the system reached its peak current).

To reduce power consumption:
- try reducing maximum power to be able to use a PicoPSU power supply; which is much more efficient at lower loads than conventional power supplies.
- opt for low-power harddrives; try to retire older 7200rpm disks which use a mighty 8W idle each and 30W spinup; use very very modern 7200rpm <5W or newest 5400rpm greens < 3W.
- opting for 2,5" HDDs can reduce idle power consumption gently, though is much more expensive solution per gigabyte
- consider powering down your HDDs, or using Wake-On-LAN to turn the PC off when not in use (at night?)

 

[Mark R]

Wow, thanks guys! I am always amazed at the helpfulness of the Anandtech community.

So instead of RAID 1, rsync or some other basic backup or cloning to another local drive is just as good?

Basically, what it sounds like is that RAID-5 for my PC is the way to go and either ZFS if I can afford it or RAID-1 + backup for the backend.
... or I could cheat and get a Drobo... (yuck)

A backup is better than RAID for security - especially, if the backup is on a different computer system.

A backup will protect you from:
Human error (e.g. deleting a file/folder by mistake, or making major revisions and then deciding you don't like them, etc.)
Virus or malware attack
Computer malfunction (e.g. faulty hard drive contrller which leads to severe file system or data corruption, overclock gone bad, etc.)

RAID can't protect against any of those things. It protects only against hard drive failure. What RAID does that a backup can't - is keep you running without a break.

E.g. your hard drive dies when you have an urgent deadline approaching. You have a backup - but it's still going to be a PITA to restore from your backup, and you may not be able to do any useful work for several hours (depending on your backup solution). RAID, however, would keep your system running.

I don't run RAID on my main system - I don't think it's really worth it. If a hard drive dies, then I just deal with it and restore from backup. If I have a deadline - well, I've got an HD image - or if it's a more serious problem - I've got a laptop which can step into the void. I have a NAS which serves as backup - and have an hourly backup script on my main PC/laptop which backs up all user data (with full version history) to the NAS. The NAS runs RAID (to protect against hardware failure), and runs a regular snapshot script that archives the backups to a partition that is not network accessible (to protect against malware attack).

Ideally, you shouldn't choose a drive in the same machine as a backup - it's still vulnerable to the same power fluctuations, CPU/hardware malfunctions, virus attack, etc.

 

[RebateMonger]

I don't run RAID on my main system - I don't think it's really worth it. If a hard drive dies, then I just deal with it and restore from backup. If I have a deadline - well, I've got an HD image - or if it's a more serious problem - I've got a laptop which can step into the void.

That's what I do, too.

I ran RAID 1 on my previous SBS 2003 server, but I never actually had a disk die on it. The failures were corruption of SQL and Exchange databases and the easiest way to recover from those was a full system restore from backups. I have a new RAID card for my new Hyper-V server, but I haven't bothered installing it. It's more work to install the RAID and migrate to it than it'd be to just restore from backups if I have a disk failure.

My "working" desktops haven't suffered a hard drive failure in years, either, and I abandoned my plans to run RAID 1 on my main desktop. I have had system corruption caused by software installation failures or driver issues, but it only takes an hour or two to do a full system restore from my Windows Home Server backups.

I DO use RAID 1 on my clients' servers. Downtime for businesses is expensive.

 

[WooDaddy]

My "working" desktops haven't suffered a hard drive failure in years, either, and I abandoned my plans to run RAID 1 on my main desktop. I have had system corruption caused by software installation failures or driver issues, but it only takes an hour or two to do a full system restore from my Windows Home Server backups.

I'm the past 10 years I've only had 3 HDs fail on me and these were desktop PCs. I've had a recent scare which has got me thinking about RAID-1 but I agree with what everyone else is saying here; RAID != Backup. I'm not truly running a 24/7 operation so I probably won't reach MTBF limit for a long time.

sub.mesa: That link is eye opening. The Atom 330 does have a higher TDP than the D510 but to see a Core i5 that low is astonishing. I might have to consider one now. Thanks for making this tougher to chose a CPU and architecture now 🙂 I also just recently learned about IMPI which is pretty intriguing to me. Necessary? Maybe not, but one of my goals is to somewhat blur the lines between IT server tools and embedded software development. BTW, I'll probably Google it, but do you have any links that describe that syncing process you're describing?

At the end of the day, I think I'll have a pretty safe and reliable backup and fault tolerant system. I perform windows backups monthly (which I need to increase) and I recently got a Bluray burner which I use to backup my VMs (that's the MOST important thing to me). It's more so figuring out where to store everything, learn a bit more about server technology all while having a fault tolerant, minimally complex architecture.

Thanks again!