Slashdot Mirror
Mirroring Controllers - What have been Your Experiences?
Today's installment is a lengthy (but hopefully informative) piece on mirroring controllers. Ever had weird problems with a FastTrack TX2000? Are you curious how well HighPoints RocketRAID boards are really supported? Ever wondered which controller gives you the best performance for every dollar spent? In true Slashdot tradition, we're taking the issue and throwing it out to you, the readers. Futurepower(R) is willing to start us off with a wealth of information on his experiences, and I'm hoping a few of you are willing to do the same.
Futurepower(R) asks: "What experiences, both good and bad, have you had with mirroring controllers? Are there manufacturers I haven't found?
For those who are interested but don't have experience, mirroring controllers (RAID 1) provide several advantages:
ECS (EliteGroup) has made 11 motherboards with on-board RAID mirroring controllers. One of them, the P4VXAD, has a Promise controller and costs about $52. What's that about? Employees at Fry's tell me that ECS has the poorest quality control of any of the motherboard manufacturers that they sell. To me, ECS motherboards seem to have surprisingly high quality. However, we have only tested three, and only one extensively.
Silicon Image makes, or made, their 0680 RAID chipset, and Koutech Systems sells the IOFLEX-Pir133 using that chipset. It costs $25 retail. I tried a Koutech card and had a lot of trouble with it, even after updating the BIOS. I talked to an application engineer at Silicon Image and a manager there. I was told that SI bought the technology from another company, and apparently it is no longer supported, doesn't work well, and the company does not intend to put more money into it. I found that the Koutech card corrupts files. The card gives a Stop 07B error when going into the Windows XP Recovery Console, unless the driver is loaded by hand, every time. I lost a lot of time with a product that apparently should not be sold.
My experience with mirroring controllers is mostly with those from Promise Technology. I've been using Promise mirroring controllers since they began making them.
The good about Promise:
The Acard controller supports SuSE, Red Hat, Caldera, and Turbo Linux, it says. But remember, Froogle found only one vendor.
HighPoint says they support Linux: "Linux Red Hat 7.3 & 8.0 (Software RAID Only)". This apparently means the card does not support Linux at all, since Linux has software RAID built in.
So, that's the extent of my knowledge and experience. Can you provide further insight?"
For those who are interested but don't have experience, mirroring controllers (RAID 1) provide several advantages:
- They prevent data loss when a hard drive fails. The other drive in the mirror takes over seamlessly.
- Reading of data is considerably faster since the controller reads the data from the drive that has a head closest to the data.
- You always have a full hard disk backup that you can pull from your system at any time, such as before installing new software.
- You can clone a Windows XP drive using the RAID card, and you will have a bootable copy. This is valuable, since the Windows XP file system cannot copy all of its own files. (Don't flame this; it has been verified many times by Microsoft employees, who often suggest using the third-party tools.)
- Promise Technology's FastTrak TX2000 is available for about $85 delivered.
- HighPoint's RocketRAID 133 costs about $80.
- ACARD Technology's AEC-6880 costs about about $85. Froogle found only one vendor. That's scary.
- Adaptec makes the ATA RAID 1200A, which is available for about $59 before delivery charges.
ECS (EliteGroup) has made 11 motherboards with on-board RAID mirroring controllers. One of them, the P4VXAD, has a Promise controller and costs about $52. What's that about? Employees at Fry's tell me that ECS has the poorest quality control of any of the motherboard manufacturers that they sell. To me, ECS motherboards seem to have surprisingly high quality. However, we have only tested three, and only one extensively.
Silicon Image makes, or made, their 0680 RAID chipset, and Koutech Systems sells the IOFLEX-Pir133 using that chipset. It costs $25 retail. I tried a Koutech card and had a lot of trouble with it, even after updating the BIOS. I talked to an application engineer at Silicon Image and a manager there. I was told that SI bought the technology from another company, and apparently it is no longer supported, doesn't work well, and the company does not intend to put more money into it. I found that the Koutech card corrupts files. The card gives a Stop 07B error when going into the Windows XP Recovery Console, unless the driver is loaded by hand, every time. I lost a lot of time with a product that apparently should not be sold.
My experience with mirroring controllers is mostly with those from Promise Technology. I've been using Promise mirroring controllers since they began making them.
The good about Promise:
- Promise has been selling RAID 1 controllers a long time. They work.
- Promise controllers can clone a hard drive quickly.
- Now some motherboards have Promise mirroring chipsets. You can get the entire motherboard with the Promise controller on the board, for maybe $50 more than the Promise controller card alone.
- Linux drivers are available. The web site says, "Windows XP/2000/NT4/Me/9x; Novell NetWare 4.1x/5.x; RedHat Linux 7.0/7.1/7.2; TurboLinux Server 6.5; TurboLinux Workstation 7; SuSE Linux 7.2; OpenLinux 3.1" Does that mean that later versions of Linux can't use this card, or did Promise forget to update the brochure PDF file?
- Promise controllers work fine under DOS, but there is no error reporting if for some reason the mirror breaks.
- Promise mirroring controllers have a software feature called "sychronization". I've asked many times over the years why it is necessary, since mirrored hard drives should be synchronized 100% of the time. I've never gotten an answer. Recently I've been told by Promise technical support people not to use synchronization, since it has caused problems. It sounds like some technical problem is being hidden.
- Promise does not support their oldest mirroring controllers under Windows XP. This is a problem since there are many business computers that are used for data entry. A Pentium II is as fast as is necessary. Windows 98 is stable with only one program running. Now those computers need to be converted to Windows XP, since Microsoft has declared that its operating systems have a curious quality: They die. (According to Microsoft, it doesn't matter that at least 100,000,000 people are using Windows 98 worldwide, it came to the end of its life on "30-Jun-2003".)
So, it is necessary to buy another controller for old data entry systems. Notice that Microsoft and Promise could decide to play this game again, and I would like to avoid the second round of buying and installing even another controller. I'd like to find a company that continues to support its products.
The speed of computers used for data entry does not matter, but the security of the data does. Hard drive failures are becoming rare, but a hard drive failure can cause a lot of problems on a data entry computer, so mirroring is required. - Some Promise controllers, especially those on motherboards, take a long time to boot. Dots crawl across the screen even if no drives are connected to the controller. Is keeping the Promise name on the screen a time-wasting sales message from Promise? Recently Promise released a BIOS upgrade for some of its cards that reduced the dot-crawling time. However, there is apparently no upgrade for Promise controllers on motherboards.
- About 2 months after I reported problems, Executive Software said they found a bug in their Diskeeper defragmentation software that might cause data corruption when used with Promise controllers. They said everyone using Diskeeper should upgrade to the new, free, minor version. I've seen no problems since then.
- Promise Technology's sales literature can be disgusting. This is the second sentence in Promise's description of the FastTrak TX2000: "The FastTrak TX2000 ATA RAID card supports Ultra ATA/133 drives to rock workstations and boost small (or large) office servers like never before." To me, this is obviously written by someone who knows nothing about the product and doesn't care.
- I find the abundant use of PDF files and unnecessary JavaScript on Promise Technology's web site annoying.
The Acard controller supports SuSE, Red Hat, Caldera, and Turbo Linux, it says. But remember, Froogle found only one vendor.
HighPoint says they support Linux: "Linux Red Hat 7.3 & 8.0 (Software RAID Only)". This apparently means the card does not support Linux at all, since Linux has software RAID built in.
So, that's the extent of my knowledge and experience. Can you provide further insight?"
look
That's all I'm sayin...
Can you spell 3ware?
They do more than just mirroring, and aren't cheap, but if you wants the quality, you gots ta pay the piper.
Drivers are in the Linux kernel, and have been for some time. ATA or S-ATA versions available.
My thoughts on the Promise FasTrack controllers:
We use 'em at work. On the Windows 2000 side, they come with decent management software and for the most part, are relatively reliable.
However, they are far from perfect. I've had several W2K servers blue screen when doing a hot-swap. Joy.
The FreeBSD drivers are bloody stable as hell. No complaints.
The Linux drivers provided by Promise are, IMHO, a POS. Pain to compile. No management software. Diagnositics are limited. As a result, I'd go with a different IDE controller card if you want it for Linux. YMMV.
Not considerably. Faster, yes, but nothing like Raid 0.
Raid 1 isn't striped, a la, your data blocks alternate sequentially between drive 0 and drive 1. Raid 1 basically issues a read request to both drives and uses the data returned from whichever drive responds first. As you said, the one whose head was closer to the data.
Contrast that with Raid 0 (or Raid 10), where blocks are read in parallel (block N from drive A, block N+1 from drive B, etc.) That parallelism leads to close to 2x read performance (on sequential reads). Versus maybe 1.1x performance improvement on Raid 1.
I have used several Promise FastTrak RAID controllers and have had varied results, but all in the "crappy" range. Their drivers are proprietary, so you have to stick with the pre-compiled kernel module. This means you also have to stick with a pre-compiled distro kernel so the symbols match. Promise has always been several kernel versions behind, so if there is a kernel security upgrade, you have two choices: 1. patch the kernel and break the Promise card or 2. Leave a vulnerable computer up and running. You can generally force a promise module into a non-matching kernel, but I've always been hesitant to do that.
I have switched every Promise installation to 3ware cards because of this. They are open source drivers, very current, and perform well. Their tools run as a web daemon under Linux so you can check status/reconfig on the fly. Really amazing Linux support, and a reasonable price... (and no, I'm not affiliated...:) )
"Just another damned fool idealistic crusader..."
Mylex controllers work very well for higher end SCSI workstations and servers. I've used these in a few systems and they are very easy to setup in both linux and windows, offer great performance, and well supported. A little on the expensive side though..
The purpose of a raid card is improved reliability. Thus, you shouldn't try to cut corners on quality control to save a few bucks. After all, you're already springing for two drives.
I have had very pleasant experiences with 3ware controllers, both under linux and Win32. Currently I have a 6400 running under Suse 7.3, and three 7400s running under RedHat 8. With some hotswap drive bays, you can even unplug a drive with the system running.
They might cost a little more, but they're widely used under more grueling conditions than the more dirt cheap designs. Also, if you're simply doing raid 1, you can use one of their previous generation cards with no performance penalty, and save a bunch of bucks.
I use a low-end 3ware raid controller doing raid1 (the 7000-2 product). It has worked great so far under linux for me.s p
http://www.3ware.com/products/parallel_ata.a
I believe it's hardware based, so it will cost a little more than a cheaper software based card(like twice as much-- around $120). I don't have the experience with a drive failing, but setting it up was painless. All you have to do is like press a key during boottime and it's like partitioning a harddrive, but easier.
I use 3ware cards and like them. The drivers are GPL unlike the Promise controllers which, if I recall correctly, are binary only.
Prevent email address forgery. Publish SPF records for y
IDE mirrors are theoretically wonderful, but don't base your entire recovery plan on them. Here's the story from about a month ago... Small webserver running on a Soyo DRAGON plus MB (Promise IDE Mirror built in), 1700XP CPU, 2 60GB in RAID1, Gentoo, etc. I set the system up in January of '02, and it took a digger this past July. Down hard, with a HD failure. When I went to check the contents of the hard drives, I discovered that one was toast (physically defective), and the other had a fully functional image of my system... from January of '03! The mirror had come appart six months ago with no errors, warnings, notifications, or anything else. I'd even rebooted the machine a few times over the last six months (new kernels, etc) and it never glitched or beeped at me. Nothing amis in the log files. No warning.
I have an SX4000, and my main problem with it is that they don't keep up on Linux support. They only recently released Redhat 8/9 drivers. For the longest time I've been forced to use Redhat 7.3 in order to use the hardware RAID capabilities. Granted, they did actually release the drivers, but back in January (when I got the card), Redhat 8 had been out for some time. When Redhat comes out with a 2.6 kernel version, I'm willing to bet you'll be waiting a year before you can upgrade. But on the upswing, hey, it does work.
In addition, I really wish they would offer wider support for their chipset in regards to Linux, as you're locked into using specific versions of Redhat as it stands now. I would really like to get Slack on this machine.
Specifically, the RocketRaid 133 (based on the HPT372A chipset). The card runs quite well under Linux. The Linux driver from HighPoint is quite good (sadly, only partially open source) and provides a /proc interface. (Don't worry, it compiles just fine in any kernel.) Sadly, I cannot show you output from the interface because Slashdot refuses to let me post it (citing junk characters). Stupid Taco. The interface also allows you to issue commands to the controller without rebooting, but documentation is poor. The BIOS utility is also quite good.
Anyway, the device hosts two RAID-1 arrays, one with 2 80Gb Seagates (ST380021A) and the other with 2 200Gb Maxtors (6Y200P0). They appear as SCSI devices. I have tested the mirroring and I am mostly satisfied.
Basically, I simulated a failure on one disk (removing it then performing some work on the other). When I reattached the drive, the card recognized the "failure" immediately and wanted to build the array. For my first test, I let the BIOS do exactly that. Took a very long time, but the mirror was recreated successfully and there were no problems (I tested by removing the first disk and trying again--the mirror was good). The second test was letting the driver do it after the machine had booted. This was a dismal failure. The card does NOT like rebuilding the mirror once the system is running.
Performance is quite good. Even though this is not HighPoint's latest offering, I am still quite impressed. I don't have any hard benchmarks, but I can post some later if you'd like.
These two arrays are accessed by many machines in my home network over NFS and by on average 5 users logged in remotely. They serve games, web pages, and my software, movie, and music archive. These arrays take a decent amount of stress, but nothing severe.
I'll post more in this thread if I think of anything. I'll answer any questions about the card's performance if you have any.
Join Tor today!
This is a huge topic, and I've only ever looked into part of it in depth, but:
Remember that for most motherboards, their RAID implementations are going to be software based, and not hardware. This makes for a cheaper chipset in general, and since the CPU speed is fairly high now the performance hit you experience is pretty nominal compared to the performance of a hardware based controller. The only REAL exception to this is for RAID 5 controllers - RAID 5 uses an XOR system which is really punishing on a CPU - hardware (i.e. add-in cards) makes a huge difference here.
All things considered, bear in mind that one of the reasons for RAID is data integrity (depending on mode, obviously) - do you want to risk your data by saving $50 on a cheap board? Where I live, the hardware place I frequent has a fairly large (10+ on staff at any one time) team of techs assembling custom orders and doing installs of purchases if their customers want them. These guys are usually the ones fixing/replacing problem MB's, and they'd be the best to talk to on quality - most reviews of MB's nowadays tend to focus on features and performance - long term reliability is something you've got to find out from others or on your own...
-- "We are all in the gutter, but some of us are looking at the stars" [Oscar Wilde]
Not offtopic (I hope), given the original poster's minor slam on ECS.
/. Wow.
I've been nothing but plesantly surprised by the ECS K7S5A/Athlon XP combo, because:
1. SiS used to suck as a chipset manufacturer, and
2. ECS used to have REALLY bad quality control.
The surprise? SiS has one of the most stable chipsets anywhere (not the best performing, but performance is irrelevant if the drivers are wonky). Add to that the fact that I've worked with about 10 boards (all bought at different times), and I've yet to have a single problem with them under Linux or Windows (in fact, the SiS chipset seems to perform better under kernel 2.4.x with the SiS chipset than even Via). I've standardized on that combo at home and with the boards I've replaced at work.
Summary: ECS has come a long way, and SiS even farther in the last couple of years. What the boards lack in performance, they more than make up for in stability.
Geez. I never thought I'd be an SiS advocate on
"Time flies like an arrow; fruit flies like a banana." --Groucho Marx
I haven't played with RAID's other than RAID5 much, but here's my experiences. Maybe they'll help your opinion.
:)
/proc/mdstat .
I've used several different external RAID5 solutions, including the Promise TX8000 and Radion arrays (and a bunch of no-name brands too). Those all worked beautifully, and survived failures without any significant problems.
I've worked with four internal hardware raid solutions. Two were the Adaptec AA-130 cards years ago, and the most recent two were Promise SuperTrak SX6000. One belonged to a friend, and one went in my machine.
My friend had several problems with his. I'll leave his story at that.
I put together a nice dual processor machine for a large photo archive site. It needed lots of space, which is why we needed the RAID5. It also needed to take up as little space as possible, so we opted for putting 6 drives in a 2u case with the SX6000. The card was incompatable with the newer chipset of the motherboard. It took two weeks of daily calls to tech support before we gave up. A week later, they released a firmware update which addressed this problem. The SX6000 doesn't handle heavy read or read/write traffic very well. When we made it an active web site, the server would crash very frequently with errors about the array. If we kept the traffic slow, it worked fine, but that what we wanted from this machine. This array solution proved to be non-functional for us, so we made the machine a backup machine, so now we have like 600Gb of storage space to back servers up to.
I've heard a lot of good things about 3ware, but haven't tried one myself yet.
I've been experementing with Linux's software RAID's. I've used RAID1 and RAID5, and they both work great. I've had a drive fail on two so far, but for the number of drives we use, that's acceptable. They rebuilt fine sticking in a new drive, with very little performance hit. I do like that Linux gives decent statistics in
You seem to want multiple platforms, so I guess Linux software RAID isn't much of a solution for you.
I can simply warn against the Promise SX6000. It should have been a good card, with Promise's reputation, but I was severly disappointed.
Serious? Seriousness is well above my pay grade.
Everyone knows Linux is quite a bit more stable than Windows. I really don't have any need to do this kind of redundant replication.
/" as root on your Linux box, again,
kiss your data goodbye regardless of
mirroring.
Assuming you didn't mean that as a troll...
HDD mirroring generally provides protection from *hardware* failure, not software glitches. If Windows suddenly decides to overwrite your porn collection, a RAID controller will faithfully do it on both drives. If you accidentally do an "rm -rf
As you can see, these controllers cost from $59 to $85. Last weekend Fry's was selling a retail boxed 1800 MHz Athlon XP processor with fan included and an ECS motherboard for $59 total. So $80 seems like a lot for a little card with one chip and a flash ROM.
Damn, we paid about $500 for an UltraSCSI 320 RAID controller and you think 80 bucks is expensive? And each one of the 75 GB drives cost $550, ouch!
I guess you allready figured out that I work for the govn't. 8) It's not the same planet as the private sector.
Think of the RAID card as a SCSI card, regardless of what it does from the board itself to the drives. You need a driver so that the kernel can talk to the board (just like for example any Adaptec SCSI card...).
Hope that helps.
The mirroring/stiping is actually done by the driver.
What's the point of that? I've always looked at hardware RAID with a sideways glance because of designs like that. None of these 'low-end' cards really advertise the fact that they are crippled in that way, either.
Wouldn't it be more redundant and just as fast to have two regular IDE controllers and just use the RAID capability of the OS? I always thought that when the drive goes, there's a fair probability that it will take the controller down with it.
If your OS doesn't have RAID built-in by now, you're not going to find drivers for it anyways.
"I assumed blithely that there were no elves out there in the darkness"
I have had a some experience with Promise cards, namely the Fastrack100 raid series, on the Win32 platform.
:(
:)
Generally they have performed reasonably well, with little hassle. However, there have been some major exceptions/issues:
* BIOS and driver upgrade -- BIOS upgraded A-OK, but the driver achitecture changed slightly between revisions (ie. one dll disappeared) and Win2K just refused to load the new driver (kept asking for a dll that no longer existed), or revert to the old driver. Joy. Spent the weekend rebuilding the server. Promise technical support was useless.
* I am yet to have Norton Ghost work properly with my RAID mirrors. The copy gets to about 99% and dies. Googling hasn't helped much either. Thus to do a ghost I have to disconnect one of the drives from the Promise and stick it on the regular IDE ports. It generally helps to remove the Promise card entirely so it doesn't throw a wobbly about the mirror being broken every time the box is reset. Joy. Mucking around in cases is fun
* The manual is generally poorly written. Explanations are unclear, and important information can be burried.
Other than that, go for it
Previously on slashdot there was a story about an article from The Tech Report that is really helpful regarding making a decision about IDE RAID card price/performance. From examining the article, it is obvious that 3Ware comes in with a better card for the money.
No, I do know what I'm talking about. Try it on your machine, if you are so inclined.
;) )
NT Workstation, 2k Pro, and XP Pro all support dynamic disks but *not* fault tolerance (defined as RAID 1 or RAID 5). They do support RAID 0, however.
The disk management interface *can* create a fault tollerant set from a workstation OS on a remote machine running NT server, 2K server, etc. That's why it's in the help file.
You can hack the dynamic disk stuff to make it think it's running a server OS (same executable images between versions) at your own discression (With the accompanying threat of trained MS attack newts killing you and everybody you care about for such a flagrant violation of the EULA.
Gentoo Sucks
Intel's ICH5R south hub seems to provide good RAID support and is lots cheaper than the PCI cards./ iaa_raid /
w thread.php?s= &threadid=8156&highlight=raptor
http://support.intel.com/support/chipsets
Sadly no Linux drivers yet, but the performance seems better than other cards due to no PCI-bus limitation:
http://forums.overclockers.co.nz/sho
Ive been using this card 24-7 in an dual Athlon system since December and its been great. I chose to do RAID 10 with 4 drives. Rebuilds do kinda suck, but the 3ware desktop manager allows you to adjust the parameters to make general use acceptable. Probably wouldnt want to spool any video while its doing this, but everything else is ok. I believe I paid about $300 then, newegg has for $275. It also does raid 5. Keep in mind they have a newer version out said to be faster. In anyevent, anybody considering raid should read this
Quite true. The Promise FastTrack is just a glorified IDE controller. But there is one very important perk: It has an onboard BIOS that understands RAID arrays and can build an array before an OS is even installed. Kinda like a built in recue disk.
Unfortunately for Promise, they are almost as stuck-up as TI when it comes to driver support on Linux. They have a few "binary drivers" that are compiled from proprietary source and only work with very specific kernel builds (namely those that shipped with a few major commercial distros).
A year or two ago I sent numerous emails to Promise's tech support asking when they would release their driver source. The answer I eventually got was basically, "whenever the hell we feel like it." So, the Promise cards at my work were resticted to use in Windoze machines or storage in scrap boxes until recently.
Recently things have improved drastically for FastTrack under Linux. Open souce drivers have finally been reverse-engineered for the controller, and it is now possible to run the controller in RAID mode reliably under the latest Linux kernels. (although it still takes a little voodoo to get the driver working.)
Because Promise RAID controllers are built into many new motherboards, I see no real reason not to use the RAID function under Linux. On the other hand, if you don't already have a controller handy, I would recommend just using the Linux built-in software RAID.
One other nice thing about the Promise RAID controllers is that if run in a mirrored RAID configuration, either of the drives can be removed from the controller and placed in a computer without a RAID controller, and all of the data remains accessable.
Just a side note:
Don't confuse the Promise FastTrack line of RAID controllers with their line of UltraATA controllers.
I've always wondered why you need drivers for RAID...
AFAIK you build your raid array for say mirroring using the bios of your raid chipset (promise for example). Then it makes sure that any write to hda is actually also a write to hdb. So I think it's purely a bios issue and nothing for the OS to do...
NT and Linux don't use the BIOS at all once the kernel has loaded into memory. The BIOS is only used for booting off the controller. It's also probably the same on OS/2, BeOS and *BSD - Win9x was a bit of a hybrid case with both protected mode and real mode stuff.
IDE users of NT have had that hidden from them a little by having a basic ATA driver bundled with the OS that works with nearly everything. SCSI NT users have always had to pick drivers for their adapters though.
I work as a sysadmin, and I am also a developer for Gentoo Linux.
Thru my career, i've used a lot of different cards.
My recommendation is solidly on the 3ware for truely solid hardware. It's is practically the only pure hardware RAID solution available. It shows up as a SCSI device to the system, and your system just needs to know that, not that 2 IDE devices are connected.
I've had horrible times trying to make the promise cards, esp the really new ones found on Tyan motherboards work, there is practically no support at all for them in linux (binary driver is no good since you can't even see the hard drives to install onto them).
The highpoint units are very good budget units. They have gone thru plenty of testing, and ship a good product, including some very nice UDMA/133 cables.
I have been in direct contact with the highpoint driver development team for the last several months helping them iron out bugs with their v2.10 linux driver, which finally contains RAID5 support now. I specificly helped them clean it up so that they could release the majority of the source code (there is a small binary part that appears to contain RAID checksumming routines and some properiatary card interface calls). There is also a largish binary only gui config tool, that I haven't personally used as they can't compile a version for my non-standard glibc setup.
If you have the money to spend, and just need mirroring, I'd put my full recommendation behind the 3ware card.
However, if you want a more budget option, or want to go further than 2 disk mirror/stripe and don't have a PCI-X slot, the highpoint is a good choice.
if you have PCI-X, go right and buy the higher 3ware models, you will not be sorry whatsoever.
I am currently running 8x 250gb maxtor drives on a highpoint controller at home for my personal storage unit, as RAID5, providing roughly 1.75tb of usable space.
ICQ# : 30269588
"I used to be an idealist, but I got mugged by reality."
As such, there are three ways of getting them working with Linux:
Use the Promise RAID driver
Use the Linux ATARAID driver
Use the Linux md RAID implementation
The first of those is handy if you're dual-booting with a RAIDed Windows installation and want everything to work. The disadvantage is that you'll be limited to certain kernel versions as other posters have already noted.
The second option also allows interoperability with other OSs RAIDed on the same drives, but because the drivers aren't written by Promise, there may be some gotchas. The advantage is that ATARAID comes as standard with all Linux kernels, so you'll never be forced to lag behind through lack of driver availability.
The third option is probably the most stable and convenient, as long as you don't require another OS to use the Promise RAID setup (md has not been ported to Windows, as far as I know!)
I chose the md option as although I do dual-boot with Windows, I don't have any important information stored there. Note also, that because all the smarts are done in software, there's no inherent performance overhead in using md over Promise's driver.
--
I know it's not supposed to be a low-end RAID-card (however, it IS :>) like the mentioned ones, but probably you also considered this one. Don't even think about it. I shared my experiences on StorageReview, click here to read it.
Free your mind!
As it happens last week I got a machine with a Promise Fasttrak TX 2000 card in it. I wanted to mirror two disks.
Unfortunately, I discovered thatthe promise drivers are proprietary and that the drivers contain all the smarts - this is *SOFTWARE RAID*, just with a BIOS control screen to configure it.
I spent a couple of days trying to get this working on Debian - and I'm no linux slouch, but I was beat. In the end I got it to recognise the controller and used the kernel's standard disk mirroring - which some people claim gives better performance than the promise controllers anyhow.
Basically, these cards are pointless if you are a linux user. The only reason I can see to use them is if you want the same drives used as mirrors for both Windows and Linux on a dual-boot system. Otherwise, leave well alone.
I've been using Promise controllers for a while now, with varied results based on method of implementation. I can't say that I'm extremely happy with them, but I've seen a lot of old info posted so, for the record, here's the deal:
/dev/hd? by default, and using them is guaranteed to destroy your array. I've also had some random kernel panics, and that doesn't fly for an HA system...
Promise has not updated their literature in a long time. They've release some source for the driver so you can run a custom kernel and the distro of your choice. You're not dependent on Red Hat or SuSE, but the binaries are still available.
Building the driver is not difficult, but you do have to edit the Makefile to fit your system. They don't provide much help, but I think most people running RAID in linux on something other than Red Hat or SuSE are going to be comfortable enough to do it.
Booting from these things suck. If you don't mind building the driver and making an initrd image to preload it, you're on your way. You should also be very comfortable with your bootloader and go over some of the linux IDE and SCSI boot parameters. If you're all of the above, it's still not a good time...
I've had some problems with the kernel support for the 20268R. Works on and off it seems. When it doesn't you have to rebuild the array and hope you didn't loose all of your data. Also, you still have access to both discs via
Now for my suggestion:
Since the Promise proprietary driver works, use it. Don't bother trying to boot off the array. Get a small disc for the system to boot from and run everything else off the array. If you want, get two small discs, use the card to make a mirror, and if the boot drive dies, plug in the other. That's the config I finally settled on, and I haven't had any downtime yet. If I could go back, I'd probably choose 3Ware, but everything is working fine so I won't complain...
ascii art