Since I am starting to re-rip in FLAC at an active rate now, I am running out of space on my main computer with the 200GB drive. I had planned for this and started building a server to house all my tunes. I also decided that it must have redundant storage. RAID5 has already been determined as the desired RAID level. I'm at the point now where I have to decide if I want hardware or software RAID.

I am using IDE disks. There is no question there. The only thing I need to know is if there is any compelling reason I should buy something like the Promise FastTrack SX4000 or just use what's built into Windows 2000 Advanced Server (my likely platform, but NT 4.0 is also an option).

If I go for software RAID, this would be the configuration:

• 1 - Samsung SP0602N 60GB Hard Disk. Used as the system disk connected to the motherboard's primary IDE controller.
• 3 - Samsung SP01604N 160GB Hard Disks. These three drives would be combined in a RAID5 array to achieve 320GB of space. I chose Samsung drives because they are quiet, cheap, and still come with a 3 year warranty.
• 3 - HighPoint Rocket133SB IDE adapters. These are cheap single channel PCI adapters with their own BIOS on board. I would like to use single channel adapters for the RAID drives because I would be able to monitor drive access with independent LEDs for each disk.

Anyone care to comment on that set up? It's going to take a lot more good points to convince me to use a RAID card over software than the DLT/DAT thing the other day.

It's going to take a lot more good points to convince me to use a RAID card over software than the DLT/DAT thing the other day.

Um, think about it Rob. This is a MICROSOFT implementation of software RAID we're talking about.

Last time I tried to use a Microsoft software RAID, I lost all the contents of the disks when one of the disks failed. Fortunately, I'd backed them up the night before onto DLT...

Also think about speed. A hardware RAID controller will be hella faster. It'll have onboard cache and such.

Hardware RAID5 will be orders of magnitude faster than software RAID5, especially in writes. Reads will be faster, but not by as much. Recovering a failed drive should be much less intensive under hardware.

Of course, all that assumes a solid RAID5 software implementation, which, from what I hear, simply does not exist in the Windows one. Every Windows admin I've ever met eschew it completely, including the rare ones who actually know what they're talking about.

You could go with an OS that has software RAID that's not failed on me in ten years or so, though: Solaris. It'll cost you $20 to download it.

Yeah, I'd agree with Tony and say you want hardware raid if you're going to do it right. Speed probably won't be too much of a factor for you, so you could probably get away with something cheap. You probably should still keep some sort of backups, however, as raid isn't a panacea, and if your raid controller crashes hard enough you can loose all your data. Then again, you could consider your CDs the backup in that unlikely turn of events.

Matthew

A hardware RAID5 card will be much faster than the software RAID option. You can also get hot-swap enclosures for hardware RAID cards which you wouldn't be able to do with software RAID.
I personally would use a 3Ware Escalade IDE RAID card but they're much more expensive than the Promise cards. The 3Ware cards have much better support under Linux and I've found them to be more reliable. I've got an 8 channel Escalade 7500-8 but it was around $500. Not sure how much the 2 channel one is though.

So what you're saying is have to be less of a cheap-ass when it comes to backups and RAID arrays?

OK, supposing I go straight for hardware RAID, can anyone recommend a good/cheap card? I don't need something with 256MB of cache, so whatever the bare minimum is and still being better than the Windows stuff would be ideal.

It's going to take a lot more good points to convince me to use a RAID card over software than the DLT/DAT thing the other day.

I guess the first three people replying to the thread bashing MS RAID was good enough.

Also think about speed. A hardware RAID controller will be hella faster. It'll have onboard cache and such.

Whilst I don't dispute that, I don't think that it's that relevant. RAID5 takes a huge hit during the write - every drive must be read in order to calculate parity. Read performance is much better. For a web server or media vault, RAID5 is ideal, for a syslog server, not so. That being said, a hardware RAID5 implementation should be significantly better than a software implementation.

This is a MICROSOFT implementation

That would be my concern.

I'll second the Escalade vote. The cheapest one you'll find that supports RAID 5 is the 7410 at just under $200. Most definitely worth the money though. In fact, you might want to consider going with a 7506-4LP for the extra $60 or so since the 74xx series is a little out of date, and the 75xx series has updated specs.

Both of those are 64-bit it seems.

Just because it's in a server case, doesn't mean it's a proper server.

Looks fine. But many dual-channel IDE controller cards also have one-LED per channel ("cable"), which is what you want, and which could save you a PCI slot or two -- many modern motherboards have only three PCI slots these days.. Think of the future (when your current mobo dies).

As for the hardware vs. software RAID: contrary to common theories, software raid has potential for the most speed (unless we're talking about 64-bit 66Mhz PCI.. which we are NOT, right?) -- especially with IDE drives. No single PCI slot bottlenecks with s/w RAID. But since you're using MS RAID s/w, all bets are off on that one.

Cheers

How often are you going to be writing to the drive? If you are mostly reading and want cheap, I would go for the software RAID. Why pay for the extra performance if you are not going to use it?

Edit: This is also assuming a non-MS software solution.

We are talking 32-bit PCI. Currently there are 4 slots available and 5 total. The other is populated by a gigabit NIC. Yeah, I'm aware 32-bit PCI can't fully support a gigabit NIC. Blah blah blah....

I have used RAID5 software in Linux before (although I didn't set up the array) and that seemed to work. Of course, the drives never died, so I am unaware how well it does during crunch time. Other than that, I have used and set up a RAID5 array with a $1300 Mylex 3 channel UW SCSI card. That thing was pretty excellent when a drive died.

So, it seems that the only problem with RAID software is when it's made by Microsoft? I can believe it.

I have no issue with RAID hardware --> just remember that you must purchase at least one spare RAID controller for use when the first one dies.. single point of failure, there.

Cheers

actually.... I have pretty much nothing to contribute to this discussion.. Allthough I have always wondered what the hell a non important server has to do with raid?

I thought that all what raid was doing, was to mirror disks. So that when 1 disk fails, disk2 takes over the job. (if for ex. your root is on disk1 and that fails the server can still operate because the root is also on disk2?)
So why would anybody want to use raid @home and use 2 disks for the same job?

Or is there any performance improvements by this? By enebaling multi read/write (?) so that the OS can read from 2 different parts off the root at the same time?
would love some pretty good explenations on this!
Got link?...=)

There are different levels of RAID. What they're talking about here is RAID 5. Much more than just mirroring disks.

There are different levels of RAID. RAID0 is commonly used by gamers and it is basically using 2 or more disks as one in order to achieve higher read/write speeds.

Mirroring one disk to the other is RAID1.

RAID5 (what I want) is the ability to take 3 drives (in my case) and use the capacity of two of them as one drive. The third drive-space stores parity information in case one of the other drives dies. If that were to happen, I could recover the data on the dead drive to a new drive and continue as if nothing happened. When we're talking about 100s of gigabytes of music that's a pain in the ass to rip, having this kind of security is desireable (IMHO).

There are probably 10 different RAID levels. 0, 1, and 5 are the most common though.

wow.... and quick reply´s too.... thanks....

The third drive-space stores parity information

That's not quite true. The parity is striped along with the real data. RAID 2, IIRC, had a dedicated parity disk, but no one uses that anymore.

hence, I said "drive-space"

The idea is you have 3 disks (480GB in my case). The capacity of two of them (320GB) is available to the user. The remaining third (160GB) is used for parity bits. Drive-space.

Hmmm. Okay. I'll let it slide.

Thanks.

And for the Novice, "Parity" in this case simply means that if any single drive of the 3 (or more in larger arrays) fails, that one drive can be replaced, and all the original information on the combined drives is still there (usually with a bit of work).

If more than one drive fails at once, you need a DLT or DVD-RAM or something else with a recent backup.

-jk

On the more advanced hardware RAID controllers it will automatically swap in a spare drive once one fails. It'll then automatically begin to rebuild the lost data in the background from the parity data spread over the other drives. This is whilst your server is still running and you'll never know anything went wrong apart from it's slower than usual.

When you've got time you replace the bad drive and reinitialise it as a hot spare. And off you go

Yeah, my current server does this. I tested it by literally pulling out one of the drives while it was running. It was like nothing had happened. Re-inserting the drive caused it to automatically begin rebuilding that drive (slowly, in the background, as the server was still fully up and running). Very nice!

It's going to take a lot more good points to convince me to use a RAID card over software than the DLT/DAT thing the other day.

Um, think about it Rob. This is a MICROSOFT implementation of software RAID we're talking about.

Actually, no it's not. The Help-About window in Disk Manager states "Dynamic disk and volume management provided to Microsoft by VERITAS Software Corporation.". And last I knew, MS and Veriats wern't getting along well, so if there was a problem, I don't see it being fixed any time soon. Took MS 3 service packs to fix an issue where a hardware RAID expansion would blow up dynamic disk layouts due to the drive info being held at the end of the hard drives.

Software (or cheep RAID controllers that simply are IDE controllers and software RAID drivers) = bad.

RAID0 is commonly used by gamers and it is basically using 2 or more disks as one in order to achieve higher read/write speeds.

And I laugh at most gamers who did this, after seeing time and time again that it ends up being slower. Why? Well the benchmarks aren't trying to do other things with the CPU during the benchmark, so it seems fine. But toss in a game loading, and all of a sudden their der cheep IDE RAID card and it's software RAID implemented via drivers chokes their system.

Raid 5, well, get a hardware controller for sure. And not just because of the dynamic disk mess, but because having to do Raid 5 calculations in software sucks. Newer Smart Array controllers (Generation 4-6) actually have IBM PowerPC processors to do dedicated Raid 5/6 calculations.

And I laugh at most gamers who did this, after seeing time and time again that it ends up being slower.

Not to mention that it doubles the chances of a catastrophic failure of a drive wiping out all your data. (Two drives with no redundancy==twice as much chance for failure.)

Okay guys.. don't get me started, but.. the only thing "cheap" about ATA (a form of "IDE", just like SCSI is..) is the price. The drives are the exact same hardware models, with different connectors and different firmware loads, but everything about them flys and is as reliable as anything else.

There's just a LOT more of them out there, which is why one hears a lot more failure stories. Not to mention that people are reluctant to make a fool of themselves by posting failure stories about hardware they paid double (or more) for..

The big differences compared with the alternatives (primarily SCSI) are:
(1) ATA is typically half the cost, or less,
(2) driver makers "differentiate" their product lines by reserving the faster spin-rate mechanisms for SCSI-only, as otherwise nobody would pay the premium.
(3) max two drives per channel on most ATA implementations, versus more on SCSI cables. But in a high end server, you really want as few drives per cable as possible, regardless of the protocol used.

Cheers

I know the drive are the same, my main point is that most PCI IDE RAID controllers in the low end are simply IDE controllers and a special BIOS/driver to enable RAID by using the CPU to do everything.

Also, the firmware does make a difference on reliability. Most Compaq hard drives that go into servers have their monitoring set to very low thresholds, so a minor hickup will be reported, and may cause the server to flag the drive as one to replace soon.

Where as I have an ATA hard drive in my media system that dosen't even do DMA transfers anymore, randomly corrupts things, but in general still somewhat works. SMART hasn't gone off once.

Other area the firmware is useful, peformance. Until recently, no manufacturer implemented tag command queuing in ATA drives.

Good points though that most drives now are pretty iidentical in their physical charistics and such.

(too tired to bother fixing the spelling mistakes and not on a browser that red underlines them)

So you are using a defective drive, one that you know is defective, and complaining about its reliablility, and therefore the reliability of all such drives?

Excuse me while I barf!

EDIT:

Tagged queuing is way overrated, though it does appear to give windows a boost of some kind (pretty much anything could help windows..). SCSI drives/controllers haven't had it for very long either, and at the same spin rate ATA drives normally outperform their SCSI brethern even without it.

I did a lot of work last year on a nice tagged-queuing controller for ATA drives, and with 7200rpm ATA mechanisms attached to it, it outperforms 10000rpm SCSI drives on a decent SCSI adapter under Windows, using tagged queuing.

But oddly enough, toggling tagged-queuing doesn't make as huge a difference with the same hardware under Linux -- the kernel already orders transactions fairly well, so unless one is using too small of a stripe size or some such thing, it's only worth a few percent, if that.

Managing a RAID in software is hardly any noticeable workload (Linux, FreeBSD) versus having a single card (bottleneck) do it all in firmware, unless one is using a ridiculously tiny stripe size (less than 128KB; 256KB is often better). Recalculating parity for RAID5 writes can be a little expensive, but if the system is that busy with I/O, then the CPU probably has enough spare cycles while it's waiting, so no big deal. Except on really big servers.

Here's a nice paper that discusses lots of RAID considerations: http://www.vinumvm.org/vinum/fullintro.html

Cheers