NAS vs SAN

[Red Falcon]

I'm kind of a noob in this area. Could someone please tell me the difference between a NAS and a SAN? I would really appreciate it.

 

[morpheus6d9]

this pic should explain everything http://en.wikipedia.org/wiki/Image:SANvsNAS.svg

 

[Lazn_Work]

NAS = Network Attached Storage, think storage accessible through the network, a share etc.

SAN = Storage Area Network, think high performance storage that your servers can share.

Your users would rarely if ever connect directly to a SAN, and your database servers would rarely if ever store their databases on a NAS.

But SANs are good places to put databases, and NASs are good shared storage for users.

 

[MrGuvernment]

what benefit does a SAN have over a raid array in a server?

or is a SAN made specifically as a high end rig with multiple databases on it?

we were considering one for work... but the money it would cost to build a high end sans to handle 3 or 4 MySQL DB's was unreal.... or at least what i thought we would need

 

[xenotype]

SAN is usually applicable for DFS (distributed file system) use, for lots of user, in excess of hundreds, usually thousands.

 

[sabregen]

...Or for a HPC shop, like us with 20 people total, mostly technical. To my boss, if it isn't FC disk, it's unreliable crap. He describes us as "geeks with $"

All of that being said, the main differences between NAS and SAN from a technical standpoint is that:

1.) NAS is file level. It has no idea what the underlying disk, block size of the file system is. It doesn't care. It's mainly used as a centrally accessible file storage system for users or user data. NAS is the kind of thing that you would implement to store people's home directories. NAS, compared to SAN, is relatively cheap.

2.) SAN is block level storage (this includes iSCSI). It has no idea what files it has (the converse of NAS), and it doesn't care. SAN is used when your data block size (from the applications being stored there) is incredibly important, and your disk is being hoit very hard (you are disk IO bound). SAN devices can customize the block size of the LUN so that they match the size of the data being generated by your application. This is IO block size efficiency, at it's best. A SAN can be tuned to have proper block size on a given LUN for a specific applcaition need, giving a 1 IO request to 1 IO fulfillment ratio.

SAN is also used when the servers that run the applications that are mission critical need to be able to share storage. Servers can share a large and fast pool of reliable storage, and the storage can be configured to give the servers just the amount of disk space that they need. If you have two servers, and they each have 4 internal (not SAN) hard drives in RAID-1, and the disks are 300GB each, each server's usable space is 600GB. One server might need 120GB of total space, and the other might need 400GB of total space. You have wasted money buying disks that you're not using. SAN doesn't have that problem. Although SAN can be very expensive, there are many cost benefits to implementing it, but it does generally depend on what you are trying to accomplish. It certainly is overkill for many organizations and needs.

In our orgranization, we exclusively use SAN. We have users home drives stored on SAN space (iSCSI) while the servers (ESX, VMs, Redhat, Fedora, etc) need the speed that Fibre Channel SAN storage can provide. It was asked what the difference is between internal server disks on a RAID controller and a SAN RAID implementation is. The first answer is cost. SAN, especially fibre channel storage, can be very expensive to deploy, depending on how much disk space you need. The second thing is complexity of implementation. Many companies do not have IT staff capable of properly configuring a SAN environment, or perhaps they cannot pay someone to manage the SAN (overhead cost) exclusively. They could of course hire a consulting firm (like us), but they do need to know how to generally configure and maintain it, or it will be even more expensive than just the initial equipment costs, warranties, support, and installation costs would be.

In addition to all of this, the real consideration to make is that many people simply do not understand why a SAN. The difference between a fully implemented and properly configured and managed SAN versus NAS or internal disk is really the robustness of the solution. SAN can usually natively snapshot LUNs on the arrays, and do block level copies that allow user access to the data before it's even done copying. Try opening a large file on a Windows server with an internal RAID controller befor the file's done copying, either from the source, or the destination. You will get an error that the file is currently in use, and likely will have issues with the file transfer itself. SAN RAID devices can change RAID levels on the fly, grow LUNs, have global hot spares, and do levels of parity calculations (RAID 6...true RAID 6, not RAID-5 DP or P+Q) in order of magnitude better than add-in RAID controllers could ever hope for. Disk drives for SAN enclosures are much more expensive than their desktop or NAS destined bretheren. I just spec'd an enclosure with 1TB SATA drives for a tiered storage array and the drives are $629/ea. You can get 1TB SATA drives on Newegg well under $200. The Enterprise class drives have been designed specifically for data center usage. Their MTBF is 1.4million hours or higher, usually. They have FDB bearings in the spindles, and tighter tolrances on the their mechanical manufacturing processes. Desktop drives are usually 500,000 hours MTBF, in comparison

NAS, by the converse, is relatively simple. They usually will use (in sub $1000 products with less than 4 disks) desktop drives, with a small PCB, utilizing an embedded processor, a proprietary OS running on a non-vollatile flash ROM of some sort. They're cheap for a reason. They're no better (and in some cases, much worse) than a NAS head machine with a decent RAID controller and some internal disk.

Hopefully I didn't bore everyone to death.

Josh

So that you don't think I am completely full of shit:
Brocade Certified Fabric Professional, Qlogic Fibre Channel Specialist, VMWare VCP, VMWare VSP, Commvault Certified Administrator, IBM System X Technical, IBM DS4000 Technical

 

[Red Falcon]

Thank you everyone for the responses. This really helps!

 

[sabregen]

glad I could do my part of it. I also forgot to mention that you can boot machines from SAN, but not NAS. For us, that's huge. We have no hard drives in any of our servers, they all boot from FC SAN.

 

[MrGuvernment]

good info sabregen ,

we have a guy at our work that was on this hitch about "we need NAS we need SAN / SAS and such, and my thought was "why ?" all he could say was cause it was good to have everything shared...

so i said, no as i didnt understand it, nor did he enlighten me as you did, and from what you have told me, my initial hunch was right, SAN is overkill for what we have or need.

 

[DrStrangeBrew]

this pic should explain everything http://en.wikipedia.org/wiki/Image:SANvsNAS.svg

That's efficiency ...probably the only wikipd page I've seen that consists of just 1 picture, yet provides useful information.

heh, & Sab had good info too - might as well post that detailed description on wikipedia for the masses

 

[priteshvarsani]

We use a SAN as a storage backend for all of our file servers. All our database servers use local disks still (We dont really need shared storage on our DB servers as our database driven apps arent exactly demanding).

A NAS is just a file server. It simply put serves files to clients wishing to access it, and manages permissions and security, and also handles authentication for the files (either local or active directory). A NAS uses a wide variey of protocols depending on how you want clients to access your file server. (i.e a windows netowork would most probably need to use SMB and if you have older UNIX servers then you may need to use NFS or something).

A SAN is a storage area network, a network of arrays of data storage which is basically like the ocean, all the storage adds up up into pools, and all of the raw storage is called the targets (the water) all of the servers that want to use storage on a SAN are called initiators (the human that needs to drink water from the ocean). Storage on a SAN actually shows up as a LOCAL hard drive on the initiating OS, which is a major advantage of using a san, SAN's come in two forms, hardware based using fiber channel, and software based using the wonderful iSCSI protocol. The main uses of SAN's are in virtual computing environments and areas where large volumes of data which require very fast storage like mail servers and database servers. a SAN is just the name for the network that stores all of these targers and initiators.

 

[unhappy_mage]

Try opening a large file on a Windows server with an internal RAID controller befor the file's done copying, either from the source, or the destination. You will get an error that the file is currently in use, and likely will have issues with the file transfer itself.

What does that have to do with SAN? If you put NTFS on your SAN device you'll have the same problem, right?

SAN RAID devices can change RAID levels on the fly, grow LUNs, have global hot spares, and do levels of parity calculations (RAID 6...true RAID 6, not RAID-5 DP or P+Q) in order of magnitude better than add-in RAID controllers could ever hope for.

Could you clarify what the difference between these three parity schemes are? I'm especially curious about the difference between "raid 6" and "P+Q".

Disk drives for SAN enclosures are much more expensive than their desktop or NAS destined bretheren. I just spec'd an enclosure with 1TB SATA drives for a tiered storage array and the drives are $629/ea. You can get 1TB SATA drives on Newegg well under $200. The Enterprise class drives have been designed specifically for data center usage. Their MTBF is 1.4million hours or higher, usually. They have FDB bearings in the spindles, and tighter tolrances on the their mechanical manufacturing processes. Desktop drives are usually 500,000 hours MTBF, in comparison

The "enterprise" sata drives are available on Newegg, too; here's a Seagate NS series disk for $240. I don't claim to understand economics, but the markup on disks in enterprise settings is fairly astounding.

 

[wasserkool]

What does that have to do with SAN? If you put NTFS on your SAN device you'll have the same problem, right?

Could you clarify what the difference between these three parity schemes are? I'm especially curious about the difference between "raid 6" and "P+Q".

The "enterprise" sata drives are available on Newegg, too; here's a Seagate NS series disk for $240. I don't claim to understand economics, but the markup on disks in enterprise settings is fairly astounding.

I can only answer the last question. He is referring to an ENCLOSURE + A 1TB Drive that c ost $600. So the enclosure alone would be around $500 because it needs be either a fibre channel or support infiniband/mini SAS connection - the usual connection in high end enterprise stuff. Lastly, organization will only use so called "SATA Enterprise" drives in non-mission critical nearline backup situations. Nothing so far even your new SSD drives will replace true SAS drives in enterprise environment all due to enterprise drives having much better parity error tolerance.

The markup on enterprise drives are well worth it because you get no questions asked 5 years exchange warranty and the drives are built at much higher tolerance level and contains many hardware/firmware features not found on your average desktop drives. Companies like seagate dedicate majority of their manpower and R&D expense towards enterprise drives and then trickle down the technology to your average desktop drives. Enterprise drives are not meant for home use anyways so the cost will not be a major concern for corporations.

 

[unhappy_mage]

I can only answer the last question. He is referring to an ENCLOSURE + A 1TB Drive that cost $600. So the enclosure alone would be around $500 because it needs be either a fibre channel or support infiniband/mini SAS connection - the usual connection in high end enterprise stuff.

Enclosures aren't always so expensive, though. The Dell MD1000 (I know, bad example) is $2500 empty, but 1TB disks for it cost $689. They're WD RE2-GPs, which cost $230 on Newegg. That's just the drive, nothing special that I can tell. But I'd bet they don't ship empty-but-usable rails with them.

Lastly, organization will only use so called "SATA Enterprise" drives in non-mission critical nearline backup situations. Nothing so far even your new SSD drives will replace true SAS drives in enterprise environment all due to enterprise drives having much better parity error tolerance.

Sure, I know. Even though you could get close to equivalent density and reliability to SAS drives by buying three times as many sata disks and using three-way mirrors...

Enterprise drives are not meant for home use anyways so the cost will not be a major concern for corporations.

True, but home users are using larger arrays than ever, and larger numbers of drives. It's only a matter of time before the demand for enterprise drives sinks down into the home user arena.

 

[Ockie]

Enclosures aren't always so expensive, though. The Dell MD1000 (I know, bad example) is $2500 empty, but 1TB disks for it cost $689. They're WD RE2-GPs, which cost $230 on Newegg. That's just the drive, nothing special that I can tell. But I'd bet they don't ship empty-but-usable rails with them.

What makes them more expensive is the warranty that is backing them and the support. EMC puts thier names on those drives and if you lost a drive, you can have a new one in a matter of 4 hours. EMC really takes care of us, we have drive replacements via Sonic Air. That is the diffrence. While you can get your drive from newegg, it does not have such a backing.

Also, you can buy the trays seperate if you wish or you can just ebay them. A lot of companies will swap out drives and they will be left with the spare tray/drive combo... but in order to secure thier data, they scrap the drive (shred it) and sell the trays.

It's not always the product you are paying for, it's a lot of the times the service/warranty/support that is included with it which makes it priceless in an enterprise enviroment.

Don't forget, the sled costs you a good 100 each. They are also not crap quality, you get what you pay for. While you can buy double of newegg's drives for the price you pay for these, companies don't see it that way, while you have no issue with redundancy, companies hosting critical applications will pay out the ass just to ensure that the chances of a drive replacement are reduced. It is also hard to get a budget passed when your justification is to buy extras or double/tripple of an item, companies would much rather pay more for something than buy a bunch of extras. If your company calls for 15 1TB drives, it's hard to tell them to buy 30 1TB drives or 18 1TB drives... because in the end, they called for "15", not 16, not 17, but 15... so at that point, you get what you need regardless of the cost... while companies appreciate the savings by using a newegg bought drive, they will not hesitate to fire you when your the array fails (think of management perspective... if they see a drive failing because you cut corners, they see it as your fault).

 

[m1abram]

Dell Enterprise also has 4 hour support. It freaked me out when I called them with a dead NIC that was on the motherboard on a PE2950 and they had a replacement board in my hands within 45 minutes. I can not get a pizza that fast!

SAN is also used in DB applications where you need shared storage, however note that the SAN does NOT handle the concurrent connections to the filesystem (it knows nothing about the filesystem). You must use a cluster aware filesystem such as OCFS if you plan to have multiple servers share the same "disk". What many SANs also provide is multipath to facilitate redundancy on the connection and load balance.

 

[wasserkool]

Enclosures aren't always so expensive, though. The Dell MD1000 (I know, bad example) is $2500 empty, but 1TB disks for it cost $689. They're WD RE2-GPs, which cost $230 on Newegg. That's just the drive, nothing special that I can tell. But I'd bet they don't ship empty-but-usable rails with them.

Sure, I know. Even though you could get close to equivalent density and reliability to SAS drives by buying three times as many sata disks and using three-way mirrors...

True, but home users are using larger arrays than ever, and larger numbers of drives. It's only a matter of time before the demand for enterprise drives sinks down into the home user arena.

Yes, enterprise can use three array system of SATA drives vs 1 array system of SAS drives but other factors starts to creep in - heat, space, power consumption and increase likelihood of drive failures. As a matter of fact, no SATA drives will touch enterprise sas arena. Event the new so called Velociraptor enterprise by WD will remain sketchy because they do not have a track record in making quality enterprise drives, whereby Seagate is the leader.

Enterprise drives will NEVER become widespread in the home because of economics and knowledge. A single 300GB SAS 15k.5 cost $500~600 and you need specialized controller to run it, which requires good knowledge of computer hardware. For example, most home users buy OEM computer from HP or dell. HP or dell will never put in a SAS controller in their home computer because of economics. You can go to the DIY route but not many home users have the knowledge. The average home user will simply not have the knowledge or money to use such drives. It might be different for enthusiast but they represent a minute market.

Sure I use a SAS drive in my workstation plus a 8 drive case SAS array because i have the need for such equipment due to the nature of work I do, but how many else would need that?

 

[sabregen]

What does that have to do with SAN? If you put NTFS on your SAN device you'll have the same problem, right?

Point taken. Typically, you would not want to put NTFS on a SAN LUN (well, I wouldn't want to) but I will concede your point here. I should have made my statement more to the point, which is:

SAN devices, being block level, can move those blocks of data inside of the array as needed, even while the file is open, or the LUN is being "snapped" (which is short for snapshot, Flashcopy, whatever you want to call it). This is accomplished by redirect on write or copy on write schemes handled by the SAN RAID contoller, and is transparent to any users of the data in question (be they servers, or people).

I did not make that very clear, so sorry for the confusion.

Could you clarify what the difference between these three parity schemes are? I'm especially curious about the difference between "raid 6" and "P+Q".

Yes, unfortunately, I can. I will try to keep this simple. RAID-5 works on distributed parity. Each of the disks in the array contain parity data, and the parity that is on any given disk is never the parity for the usable data that is on the disk. In this way, when a disk in RAID-5 fails, all of the parity for that disk is stored elsewhere. All other disks in the array are accessed to speed the rebuild time of the failed drive (once it's replaced, that is). RAID-5 looks like this:
http://upload.wikimedia.org/wikipedia/commons/f/fc/RAID5.png

RAID-6 (true RAID-6, that is) adds another level of protection to RAID-5. You see, RAID-5 is great for data protection, but it does have one point in time where you are completely screwed if it fails: the array rebuild. The larger the array, the more full the disks are, and the slower the disks, the longer the rebuild. During a RAID-5 rebuild, any other single failure will cause the entire array to fail. Rebuild should be when you have your fingers crossed, and are praying to whomever that the whole shit doesn't just go up in smoke right in front of you. RAID-6 seeks to alleviate this potential crisis by adding another set of parity into the mix. In this, RAID-6 (whether true RAID-6 or DP or P+Q) can sustain 2 failures per array. The inherent problem is that it takes longer for the rebuild to complete. In many ways, it's a catch 22, or at the very least, a compromise in speed for security...but then again, that's why companies do RAID-5 to being with...security over speed. RAID-6 just adds extra protection.

So as to not get too off track with "Why RAID-6?" let me explain in as few words as possible the differences between True RAID-6 (the way Hitachi and IBM do it), and RAID-5DP (also know as RAID-5 P+Q, RAID-6, RAID-6DP, or RAID-6P+Q...confusing huh?).

True RAID-6 involves an extra partity stripe on the disks in the RAID-6 array. It looks just like RAID-5 with an extra parity stripe on the disks. The second stripe goes in the reverse order of RAID-5 (because that just makes sense). Any single drive in the array can fail, and the array will go critical. When the single failed drive is replaced, all of the other drives (recall that they all contain parity data that is not their own) kick in on the rebuild. This helps the rebuild times significantly, as all disks are being accessed to complete the job (which is the point of any RAID, in the first place...more spindles+better performance).
True RAID-6 (like IBM or Hitachi) looks like this: http://upload.wikimedia.org/wikipedia/commons/thumb/7/70/RAID_6.svg/300px-RAID_6.svg.png

The other RAID-6 (which for the sake of simple naming convention), I will just call RAID-6/DP/P+Q (if that's simple...) uses a differrent method of distributing the parity. It does it all on two drives. Yes, you read that right. The controller for the array will physically assign two disk drives (not virtual drives, not logical drives, not LUNS...physical, tangible, disk drives that have mass) as parity drives. I'm going to let that sink in for a moment, and I think you'll understand what I am about to say a lot better...

This implementation has so many drawbacks, it's astounding. I will give an example from above to compare: A single drive failure in a normal RAID-6 array (true RAID-6, like IBM or Hitachi) would cause the array to go critical. However, unlike the normal RAID-6 (above), RAID-6/DP/P+Q only has 2 physical drives (defined as the physical parity drives by the controller) from which it is recalculating the data on the lost drive. Instead of all drives in the array assisting in the rebuild, now there are only two. The question that immediately comes to my mind is "What happens when you lose a parity drive?" Glad you asked, but I bet you can guess already...the remaining parity drive fills in all of the gaps that it can, then goes out the non-failed disks and grabs the data that it needs from them to fill in the holes in the stored parity. In this implementation will be in orders of magnitude slower than either true RAID-6, which is orders of magnitude slower than RAID-5. It's dog slow..I mean terribly, terribly, painfully slow to rebuild, but it it is much cheaper to have a controller than can just assign two disks as parity than it is to build a controller than can mathematically calculate two striped sets of parity every time a write is performed. You get what you pay for, one way or the other. True raid-6, you pay up front, RAID-6/DP/P+Q, you pay when a drive fails.
RAID-6/DP/P+Q looks like this (note that the two disks on the right, labbeled "P" and "Q" are the parity disks): http://upload.wikimedia.org/wikipedia/commons/9/98/RAID_DP.png

The "enterprise" sata drives are available on Newegg, too; here's a Seagate NS series disk for $240. I don't claim to understand economics, but the markup on disks in enterprise settings is fairly astounding.

Well, yes and no on these being enterprise drives. I have no doubt that Seagate will label the drive as such, and I'm sure that their bearings, and vibrational ratings are in line with being an enterprise drive, but here's what happens:

OEM for SAN RAID chassis designs their chassis with certain specifications in mind...drives have to have a minimum number of spindles, or have a certain temperature threshhold, or have only a certain amount of vibration, etc. Based on these chassis specifications, they select drives for each enclosure from a manufacturer, like Seagate, or whomever. They test the drives extensively in the chassis. If the drives do not reach a certain failure threshold (usually a percentage of the total), then the OEM (like LSI Logic, IBM, Sun, whomever) will qualify those drives for that chassis. This process is repeated a new chassis or drives comes into production. This is part of the cost of the drive and/or chassis on the market, the folded in R&D costs. On top of all of that, the OEM of the RAID enclosure, let's pick IBM, will actually OEM the drives out as well. They might be Fujitsu or Seagate drives, but the label on the drive is going to say IBM. The same thing happens with server RAM, and all sorts of other components. This is why the drives are so expensive.

To the poster that attempted to interpret my meaning when I said 1TB drives are $629 each from IBM...no...I really mean that just a single 1TB SATA drive, 7200RPM, that has an IBM sticker on it, and is qualified from anything starting at a DS320 to a DS4800 enclosure is $629 each...yes, 1 single 1TB drive. When it fails, you're calling IBM, and they're the ones replacing it for you, and usually within 4-8 hours. Try that with Newegg.

 

[Ockie]

Yes, enterprise can use three array system of SATA drives vs 1 array system of SAS drives but other factors starts to creep in - heat, space, power consumption and increase likelihood of drive failures. As a matter of fact, no SATA drives will touch enterprise sas arena. Event the new so called Velociraptor enterprise by WD will remain sketchy because they do not have a track record in making quality enterprise drives, whereby Seagate is the leader.

Enterprise drives will NEVER become widespread in the home because of economics and knowledge. A single 300GB SAS 15k.5 cost $500~600 and you need specialized controller to run it, which requires good knowledge of computer hardware. For example, most home users buy OEM computer from HP or dell. HP or dell will never put in a SAS controller in their home computer because of economics. You can go to the DIY route but not many home users have the knowledge. The average home user will simply not have the knowledge or money to use such drives. It might be different for enthusiast but they represent a minute market.

Sure I use a SAS drive in my workstation plus a 8 drive case SAS array because i have the need for such equipment due to the nature of work I do, but how many else would need that?

I think mage was talking about enterprise sata drives, not sas drives.

 

[sabregen]

I think mage was talking about enterprise sata drives, not sas drives.

I tihnk he went off on his own tangent, because almost none of it is directly related to the OPs thread.

 

[unhappy_mage]

The second stripe goes in the reverse order of RAID-5 (because that just makes sense).

What do you mean by "the reverse order"? Are you talking about the parity generation scheme or the placement of the parity onto disks?

The other RAID-6 uses a differrent method of distributing the parity. It does it all on two drives.

Oh. Okay. I thought you might have been talking about NetApp's RAID-DP, which is significantly different from RAID 6 and should not be lumped in with it. If I've got this straight, what you call "RAID-6/DP/P+Q" is to RAID 6 what RAID 4 is to RAID 5.

unlike the normal RAID-6 (above), RAID-6/DP/P+Q only has 2 physical drives (defined as the physical parity drives by the controller) from which it is recalculating the data on the lost drive. Instead of all drives in the array assisting in the rebuild, now there are only two.

All of the non-failed drives, including any parity disks that are left, are required to rebuild the data. Thus, all the drives "contribute" to the rebuild. The parity drives alone do not contain enough information to reconstruct a failed drive, (N-2) of an N-drive array are required.

this implementation will be in orders of magnitude slower than either true RAID-6, which is orders of magnitude slower than RAID-5./QUOTE]
I would be shocked and astounded if you could demonstrate even one order of magnitude difference between raid 5 and 6DP on the same physical set of disks. Even in base 2. Thirty or forty percent difference, sure. Not 900%.

To the poster that attempted to interpret my meaning when I said 1TB drives are $629 each from IBM...no...I really mean that just a single 1TB SATA drive, 7200RPM, that has an IBM sticker on it, and is qualified from anything starting at a DS320 to a DS4800 enclosure is $629 each...yes, 1 single 1TB drive. When it fails, you're calling IBM, and they're the ones replacing it for you, and usually within 4-8 hours. Try that with Newegg.

I understand that. But my point is that home users will look at the IBM sales page, the Newegg page, and buy from Newegg. But if IBM had a cheaper-support option for home users, they might be able to get that business.

I tihnk he went off on his own tangent, because almost none of it is directly related to the OPs thread.

Yep. But the OP's question is answered, and there were some loose ends left open. I could have started another thread and asked there, but this one already had all the required context in it.

Please, OP, if you'd rather I moved to another thread, let me know and I'll be glad to move.

 

[sabregen]

This one I am keeping short.

Point #1: I didn't look at the diagram that I posted that closely. It is my understanding (from an IBM storage camp) that Hitachi and IBM's implementation of RAID-6 differs a bit from that diagram, where the parity top to bottom would actually be as follows:

Ap, Dq
Bp, Cq
Cp, Bq
Dp, Aq

So that one parity stripe (in a graphical representation) goes from the lower left to the upper right (the p set), and the next set (the q data) goes from the top right to the bottom left.

Point #2 - You are correct in your comparison. In my opinion, DP is the way to do it wrong...which is the way most people do it, unfortunately.

Point #3 - You have me wondering if I am wrong. I may call my contacts at LSI and ask them about this.

Point #4 - I should choose my words more carefully. You are correct, a 10 fold increase in time, let alone multiple 10x increases in operational window would render the product entirely useless because the overhead would be too much to even consider implementing it. I misspoke. The time to rebuild a RAID-6DP array is significantly more than a true RAID-6 implementation. RAID-6 takes significantly longer to rebuild than a RAID-5 array.

Point #5 - duh. But I don't think any home user would buy IBM's enterprise level drives, even without the costly support. Most OEMs, like IBM, require a support contract as an additional charge to the equipment that the contract is for (at least, for upgraded service). Take out the service, and you still have extremely expensive parts in comparison to consumer level parts.

 

[MrGuvernment]

True, but home users are using larger arrays than ever, and larger numbers of drives. It's only a matter of time before the demand for enterprise drives sinks down into the home user arena.

large arrays @ home vs Enterprise, no comparison, we aren't talking a couple hard drives in raid 5 here with a corsair power supply, at least that isn't my thought.

 

[sabregen]

lol, have you seen Ockie's stash!?

 

[m1abram]

On our SAN we actually use RAID 10. It provides better fault tolerance than RAID5 and even RAID6 in most cases and provides better write performance.

We use a 10 disk RAID10 with a hot spare. This system can suffer multiple drive failure as long as the right drives fail . Actually if all the right drives fail it can suffer a 5 disk failure (good luck betting on that). However since it would require 2 drives in one of the 5 RAID1 arrays to fail at the same time that is rather unlikely. Also with a 2 drive RAID1 (which is all you are concerned with if a single drive fails in the RAID10) the rebuild time is much much faster than any RAID5 rebuild and does not effect the performance of the array as drastically as RAID5 or RAID6. Since we use our SAN for an Oracle RAC I/O performance on both writes and reads are important.

Other bonus RAID10 scales to larger sizes without reducing the fault tolerance like RAID5 and RAID6. Course the cost is more drive space.

 

[Red Falcon]

Please, OP, if you'd rather I moved to another thread, let me know and I'll be glad to move.

You're ok. Everyone here is giving logical arguments and answers. Keep em coming.

 

[sabregen]

On our SAN we actually use RAID 10. It provides better fault tolerance than RAID5 and even RAID6 in most cases and provides better write performance.

We use a 10 disk RAID10 with a hot spare. This system can suffer multiple drive failure as long as the right drives fail . Actually if all the right drives fail it can suffer a 5 disk failure (good luck betting on that). However since it would require 2 drives in one of the 5 RAID1 arrays to fail at the same time that is rather unlikely. Also with a 2 drive RAID1 (which is all you are concerned with if a single drive fails in the RAID10) the rebuild time is much much faster than any RAID5 rebuild and does not effect the performance of the array as drastically as RAID5 or RAID6. Since we use our SAN for an Oracle RAC I/O performance on both writes and reads are important.

Other bonus RAID10 scales to larger sizes without reducing the fault tolerance like RAID5 and RAID6. Course the cost is more drive space.

RAID-10 is sweet, I'll give you that. If my laptop could take 1 more drive, that's what I'd be running. Yeah, good luck getting the lucky one against the odds on all 5 proper disks failing the in the 10 disk array. I don't even know what those odds are, but they've got to be really fricking high, and not in your favor!

 

[m1abram]

RAID-10 is sweet, I'll give you that. If my laptop could take 1 more drive, that's what I'd be running. Yeah, good luck getting the lucky one against the odds on all 5 proper disks failing the in the 10 disk array. I don't even know what those odds are, but they've got to be really fricking high, and not in your favor!

Yes I know, but my main point was it *can* suffer multiple drive failures and still keep going. Your odds of having a BAD multiple drive failure decreases when the number of drives in the array increases (this is a good thing). And with rebuild time much faster than RAID5 chances of a failure during rebuild is much lower.

 

[DEX1]

I've had a SimpleTech NAS for years and am totally satisfied.

 

[wasserkool]

I've had a SimpleTech NAS for years and am totally satisfied.

Sure, that NAS is good for home users who just want to share some small network files. For people who needs the fastest throughput in read write when dealing with high-defintion video or photos, a simple one disk NAS will not suffice.

For example, my NAS can easily saturate my gigabit connection when I am doing intensive stuff.

That is why we have people here with SAS drives in arrays as their NAS.

 

[Red Falcon]

At my job, we use a SAN connected with fibre channel. If we were using a NAS, people would be sooo pissed at the slow transfers, it would be bad.

NAS = good for home

SAN = good for enterprise