A RAID that just works!

ratbat

Distinguished
Mar 20, 2009
7
0
18,510
A RAID that just works!
I run an IT support company, so I talk to clients every day who are sobbing over broken hard drives. Kind of makes you paranoid about backing everything up. My own machines are always configured in at least a mirror (RAID 0 is evil) and I have been using RAID 5 on my personal main desktop to squeeze out more performance (I know, I know, I read the thread about how it does not make much difference, but I run a lot of VMs and move a lot of data around, so there are a lot of parallel disks accesses and so the performance does make a difference).

My question? I need to get off of using motherboard RAID and get a dedicated controller. I just had a problem which convinced me of that when I had a drive failure (I think the sata cable worked lose). I fixed that and the controller is talking to the drive again, but Windows crashes on start up (I assume the driver is crapping out), so Windows never starts fully and never rebuilds the RAID. Looks like I am going to have to reinstall it and I could do without the hassle.

I need a system that just works. I need to be able to laugh at a disk failure, slip out the drive and pop in another without breaking stride. Or if the controller breaks just buy another one (even from another manufacturer) and plug it in without having to go into recovery mode. But what should I buy? There are a lot of controllers out there and I do not know how to choose. Can you help?

For now I need something that will support 3 x 1Tb Samsungs (although I think I might need to add a fourth one) all in RAID 5. For the future I like the idea of adding 3 additional SDDs in RAID 5 and moving the boot partition onto them so that I get lighting fast software load times. All I am waiting for is the price of these units to drop from the upper atmosphere (which I am hoping will be by the end of the year).

Robustness is the key thing. I want a disk failure to be a non-event that just flags itself for attention. Great benchmarks are a fine thing, but that robustness is even more important to me. The reviews I have read of controllers all focus on the raw performance numbers, none of they say what would happen if they pull out a drive in the middle of the test. I had one machine last year on a motherboard RAID that will reconstruct the RAID, but the machine was unusable (way too slow) for a day or so whilst it did it. That’s the sort of thing I am interested in. Any of you guys had drives fail on this type of controller whilst in operation? What happened?

In a dream world I would like a Drobo like solution where I could notice I was running low on storage and bang in another 2Tb drive and the system would automatically and redundantly redistribute the load. I have not gone to a Drobo because I cannot boot off of it and I do not have any software that would recover from it if it did break (I use Runtime Software’s RAID Reconstructor when I need to fix recover a RAID, but I do not think there is anything out there for a Drobo). I add this note in case my lack of knowledge of the RAID world means there is a magic bullet solution that you can tell me about that will fix these issue. A long shot I know, but worth a try.

The rig is a Core i7 overclocked at 3.8Ghz with 12Gb of RAM and Windows 7 64 bit.

What words of wisdom can you offer?
 

MRFS

Distinguished
Dec 13, 2008
1,333
0
19,360
Well, you're probably going to get as many different answers
as different people who reply here.


My colleague Roger Shih at Micro City, Inc. in San Diego:
http://www.microcityinc.com/
has been building custom servers for many years, and
he has routinely recommended Areca hardware RAID controllers
with RAID 6 arrays.

(Just be patient because his native language is Chinese,
and sometimes his English is just not perfect: he is,
nevertheless, an expert building reliable storage subsystems --
particularly for large servers.)

I would also add that, for sheer reliability and without the
extra cost of SAS drives, Western Digital's RE3 series are superb.

Be realistic and EXPECT rotating drives to fail:
WD has always been there for me with a very rapid RMA replacement:
last time, they couldn't replace a 400GB drive that failed,
so they sent me a "tire-smoking" 750GB that almost
doubled the original capacity, with much higher speed.

I WAS THRILLED!!

I would suggest that you implement a hybrid system, and
run your OS from a dedicated partition on a Western Digital
VelociRaptor, also with a 5 year factory warranty.

Now, just assume that the VR is going to fail at some point
in the future. When it does fail, you will probably have
many many hours invested in installing and tuning system
software, much more time-value than the replacement
cost of the hardware itself.

So, the obvious solution is to write regular drive image files
of your OS partition, and save them everywhere you can,
including onto a second data partition on the VR.

It's generally NOT a good idea to format the entire VR
as a single NTFS partition.

So, here are the probable scenarios:

(1) HDD with OS fails. Solution: shutdown, replace it, reboot and
execute the drive image restore task -- 15-30 minutes max.

(2) RAID 6 member disk fails. Solution: swap it out
and replace it with an identical new one -- 10-15 minutes max.

With wise component choices, your RAID 6 should support "hot swaps".

The better hardware RAID controllers come with excellent
management software: we've had a lot of success
with Highpoint's RocketRAID controllers, but then our
management requirements are rather modest.


As you can probably tell, I'm a very big fan of Occam's Razor:
the simplest solution is always THE best solution.


Good luck!

p.s. Anand has done another one of his brilliant analyses here:

http://it.anandtech.com/IT/showdoc.aspx?i=3532
(definitely worth a careful study: he's an obvious genius)


MRFS

 

Its_Me_Mario

Distinguished
Mar 19, 2009
18
0
18,510
I just learned how to set up a RAID and it took me 2 weeks. So I will not be much of use to your initial cause, but I can tell you how you might get good info on this. Go to newegg (which I am sure most people on this site are aware of) and just read through the reviews. When I am looking around for a part, I set the customer reviews to as many as I can per page, and *Ctrl F keywords to pick out the pertinent info I need. This way you get it from actual consumers, none of the manufacturers are going to tell you the bad.
 

Gatorbait

Distinguished
Jan 29, 2009
63
0
18,640
Hi Mario,

First off you should get a HW RAID HBA controller. (This mean on-board DDR Memory and will cost more)

I would suggest the number one RAID HBA supplier in the world: LSI. Why? Well, because they make their own ROC (RAID on CHIP), the RAID HBA, the RAID FW, and OS drivers. They also sell to all of the major OEMs, so their product is going to be solid and cover all of the OS varieties.

http://www.lsi.com/storage_home/products_home/internal_raid/megaraid_sas/value_line/index.html

If you have a UPS, forgo the costly RAID Battery backup module, but make sure you set the controller to write back for better performance.

I have used their product on numerous occasions. If a drive fails, you just pop in another and let it rebuild on the fly. I would absolutely suggest that the OS is not on the same array as the data (and maybe not on an array at all, but on the host using RAID 1, as most MB support RAID 0 and 1 pretty well). Why: The OS is tied to the MB and if you want to move the array to a new MB, you will need a new OS install anyway (or an OS headache converting it to the New MB... Unless you have an Identical backup).

Note: SAS/SATA 6G is coming out. The New RAID controllers are/will also be out too... I do not know if the cost will justify the future unreadability, but I though I would throw it out there.

Other RAID HBA manufacturers:
Adaptec
Highpoint
3ware
Areca

Note: I can understand why the MRFS recommended RAID manufacturer expert (whose native language is Chinese) would use Areca; they are a Taiwanese company, and thus, probably have a better Chinese users manual.

Good Luck,
Gatorbait
 

Translucency

Distinguished
Mar 26, 2009
90
0
18,630

RAID0 is not evil. ;) In fact, i would argue that a RAID0 array of 4 SSDs would be safer than a RAID5 array of 4 HDDs. But since you're looking for mass-storage and not high-performance, SSDs are ofcourse no option.

While you're looking for a 'enterprise-quality' storage solution, looking in a totally different direction you can also concentrate on a good working backup solution. Say you have a RAID-array in your main system, add to that a NAS/fileserver connected via the network that acts as backup fileserver, so in the case of failure of your data array (or temporarily loosing access to it for whatever reason) you can always fall back on your NAS fileserver. That's what i call redundancy! You access the backup-array as just another drive letter to Windows where an exact copy of your data lies.

This means you have to synchronize it, which there are several tools for. Do this in an automated fashion every night and you have a quick means of gaining access to your data should you have any problem with your internal array. It also might prevent you from dangers as viruses, filesystem corruption and accidental deletion since the backup fileserver contains data up to 24 hours old if you synchronize once every night. Using live synchronization is discouraged since it impacts both data security and performance; syncing at night when you're asleep won't impact performance at daytime.

and I have been using RAID 5 on my personal main desktop to squeeze out more performance (I know, I know, I read the thread about how it does not make much difference, but I run a lot of VMs and move a lot of data around, so there are a lot of parallel disks accesses and so the performance does make a difference).
It indeed does. :)

My question? I need to get off of using motherboard RAID and get a dedicated controller.
While i agree that onboard RAID (also called fake RAID) has many examples of bad implementations and customers with datacorruption or inaccessibility because of a software issue. However, software RAID is not inferior to hardware RAID. In some cases it is even superior. Should you use Linux or FreeBSD on the backup server i talked about, you would be able to use advanced filesystems like ZFS. This gives you the most maintenance-free reliable filesystem out there, which heals itself when it finds a problem and due to its design has many protections in place that NTFS does not offer.

While you may create an ultimate RAID5 or RAID6 array with a high-profile enterprise-class controller, you're still stuck with NTFS when you're using Windows. While you may not want to change your operating system on your primary computer, you may want to on a backup server which once setup has virtually zero maintenance - unless a disk fails ofcourse and you should physically replace it, click the new disk to be part of the array and it will re-build automatically. At neither point in this procedure will you experience any inaccessibility or serious slow-downs. In fact, due to its design, rebuilding a disk in ZFS might be very quick because it only transfers data it needs to know instead of all contents from beginning to end like RAID controllers do.

In a dream world I would like a Drobo like solution where I could notice I was running low on storage and bang in another 2Tb drive and the system would automatically and redundantly redistribute the load.
ZFS is not that far yet, but it certainly is a step closer than NTFS is. ZFS is essentially an advanced filesystem and RAID-engine in one, and can benefit from coupling these together making filesystem corruption a thing of the past. I wouldn't say its a silver bullet yet, since ZFS is still a work-in-progress; but its certainly good to investigate if it can benefit you.
 

TRENDING THREADS