Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[christopherw248]

In reading about this product and other NAS products, many of them use some form of linux.  However, the documentation is sparse as to the method of RAID.  Does this product use hardware or software RAID.  Hardware is preferable to me personally.

[MasterChiefDutra]

If you give me a day or two I will be able to look that answer up for you, I don't know off hand.

[mig]

Software RAID

[christopherw248]

If it is software RAID, I wonder what level of confidence can be placed in it.  I know that mirroring is relatively simple to do software wise and thus if there's a problem hopefully at least one of the drives has the right thing.

Drives that have been RAID 5 together are not so simple.

I've heard varying information on software RAID.

Does anyone have any experience using software RAID 5...?  Is it pretty stable...?

[sfrede]

The question of software vs hardware RAID in a device like this is an oversimplification.  What do you mean by "hardware RAID"?  That the signals to the disk are controlled by TTL logic chips or an ASIC?  Hardware RAID implementations will normally have an embedded CPU of some sort.  That CPU runs a program, performing the same sort of operations as would a driver in a general purpose computer system.  The reason hardware implementations may be considered more reliable is that they operate in a much more well-defined space.  A driver in an operating system has to cope with a much larger range of possible environmental variables - interrupts, locking issues on multi-core systems, a wide variety of hardware platforms, a huge variety of competing software, not all of which may be well-behaved, even different versions of host operating system.  All these things are controlled in RAID cards.  But to a large extent they are also controlled in dedicated purpose devices like this.  Yes, there is a version of Linux running under the covers, but the version, the software environment, the hardware, the cpu - everything is fixed so that a test suite can be much more comprehensive than in an environment where these variables aren't controlled.  Then you also have the advantages of "software" implementations.  You can make changes much more easily.  In a version of RAID on a card, if subtle bugs are discovered (and they very often will be), does the manufacturer recall all the cards?  No.  But in a software implementation, the manufacturer can issue updates, as Dlink has done here.  Another advantage of using an open source implementation as a base is that the software has been through a very wide and rigorous test process.  For my money, a device like the Dlink NAS (and others that are similar) provides both the flexibility of an upgradeable implementation with the consistency of locked down hardware and software.

[fordem]

The question of software vs hardware RAID in a device like this is an oversimplification.  What do you mean by "hardware RAID"?  That the signals to the disk are controlled by TTL logic chips or an ASIC?  Hardware RAID implementations will normally have an embedded CPU of some sort.  That CPU runs a program, performing the same sort of operations as would a driver in a general purpose computer system.  The reason hardware implementations may be considered more reliable is that they operate in a much more well-defined space.  A driver in an operating system has to cope with a much larger range of possible environmental variables - interrupts, locking issues on multi-core systems, a wide variety of hardware platforms, a huge variety of competing software, not all of which may be well-behaved, even different versions of host operating system.  All these things are controlled in RAID cards.  But to a large extent they are also controlled in dedicated purpose devices like this.  Yes, there is a version of Linux running under the covers, but the version, the software environment, the hardware, the cpu - everything is fixed so that a test suite can be much more comprehensive than in an environment where these variables aren't controlled.  Then you also have the advantages of "software" implementations.  You can make changes much more easily.  In a version of RAID on a card, if subtle bugs are discovered (and they very often will be), does the manufacturer recall all the cards?  No.  But in a software implementation, the manufacturer can issue updates, as Dlink has done here.  Another advantage of using an open source implementation as a base is that the software has been through a very wide and rigorous test process.  For my money, a device like the Dlink NAS (and others that are similar) provides both the flexibility of an upgradeable implementation with the consistency of locked down hardware and software.

What a mouthful - ever thought of breaking things into paragraphs? it would make things much easier to read.

In the event of a bug being discovered with hardware RAID, the manufacturer of that RAID card simply releases a firmware update for the card - in fact, the very first RAID card I used (back in 1993) had a very serious bug that prevented the array from completely rebuilding after a failed disk was replaced, and the fix was to upgrade the RAID card firmware.

[jswashburn]

In the event of a bug being discovered with hardware RAID, the manufacturer of that RAID card simply releases a firmware update for the card - in fact, the very first RAID card I used (back in 1993) had a very serious bug that prevented the array from completely rebuilding after a failed disk was replaced, and the fix was to upgrade the RAID card firmware.

Just to clarify for those who don't already know. Firmware = software on ROM (EEPROM/Flash Memory to be exact) which is intended only for the device itself and thus independent from your Operating system. So when you update the firmware on a device, you’re actually updating its internal device "driver".

Technically, all RAID cards are software based. The only difference is that "hardware" based RAID cards run the software from its own ROM chip and have the CPU directly on the card itself. "Software" based RAID cards are just a basic controller that sends all traffic to your main CPU which in turn does the calculations inside your operating system (Linux, Windows...etc). With a standard PC, "software" based RAID card, it will start to saturate your south-bridge chip when you start adding multiple drives to an array. With a "hardware" based RAID card, all of the calculations happen on the cards own bus and CPU. As you might imagine, software based RAID cards don't scale very well in a server environment.

I guess what I'm trying so say is this. Being that we are only using two drives in RAID 0 or 1, it's actually better to have a software based RAID system. It's far more manageable and can be patched with each firmware revision D-Link releases. Also, the type of hardware they are using isn't your standard PC motherboard configuration so I doubt there's an I/O bus saturation. Then again, I don't know what the bus width/speed specifications are so I could be wrong... Anyone at D-Link care to tell us what's exactly under the hood?

[fordem]

I'm not certain I follow your logic there ....

I can't agree that is it "better" to have a software based RAID system - it is certainly cheaper, and at the price point that these units are designed for, probably the only way to do it - with hardware RAID the price would be significantly higher.

A couple of points ....

First - we're discussing the DNS-343 here, so I believe we also have the option of RAID5.

Second - I believe that regardless of the RAID level involved, hardware RAID has advantages - there may not be a significant performance advantage when dealing with RAID0 or RAID1, but there certainly is at RAID5.  There is also a manageability issue - I have been using hardware RAID for close to fifteen years now, so called "FakeRAID" (what you describe as software based RAID cards) for almost three years, and software RAID (where the RAID functionality is handled by the OS - which is how it's done on these units) for around six years - with hardware RAID & FakeRAID, the RAID functionality is completely transparent to the operating system, it does not know how many physical drives exist.

From a maintenance point of view, replacing a failed drive is a snap with today's hardware RAID controllers - look for the orange LED, pull the drive out, put a new one in and watch it rebuild (assuming you have hot swap hardware and SATA is designed for hot swap).  FakeRAID on Windows is no different, it works exactly the same way - however - for reasons best known to themselves, the linux community seems to have chosen to shun FakeRAID (I believe that is where the somewhat derisive term was coined) preferring to handle it in the operating system.

By comparison - replacing a failed drive with software RAID can be a living hell, especially if the unit boots it's operating system from the RAID array - fortunately the D-Link DNS series of NAS deviuces boot the OS from flash, making things just a little easier.

It is my belief that many of the data loss scenarios documented on the DNS-323 (the 322 & the 343 are still too new to have developed a reputation, either good or bad) are the result of scripts that attempt to automate the drive replacement process - scripts that no doubt have been written to the best of the developers ability, but scripts that fail to take account for all of the possible scenarios that can occur in the real world.  The fact that Linux will very cheerfully assign hda (or sda) as the first disk drive it finds with no consideration for the hardware port the drive may be connected to is no doubt a contributing factor.

As you have just pointed out - firmware on a hardware RAID card is nothing more than software dedicated to the task at hand - it is no more difficult to flash the firmware on a hardware RAID card than it is to flash the firmware on one of these units.  In fact, if D-Link had chosen to use hardware RAID, then a RAID firmware flash could be incorporated in an OS firmware flash without the user being aware of it.

No, I definitely cannot agree with you - if anything software RAID - as in RAID handled by the operating system - is much more difficult to manage than either FakeRAID or hardware RAID, and has to be considered as the price we pay to keep the cost of the product where we can afford it - if you like, a necessary evil

[dlach]

Seems like the motherboard picture in the review on SmallNetBuilder.com indicates we have software RAID, the theory is supported by the very long RAID rebuild times reported.

Still it would be nice to have someone (ECM?) from DLink give us the word.

[bspvette86]

Guess it depends on your perspective.  If you are on a pc using the network share, the DNS-323 is a piece of hardware doing the raid not the OS. 

If you are on the DNS playing around at the OS level, it is Software raid...

Regards
BSPVette86