Slashdot Mirror
IDE RAID Examined
Bender writes "The Tech Report has an interesting article comparing IDE RAID controllers from four of the top manufacturers. The article serves as more than just a straight product comparison, because the author has included tests for different RAID levels and different numbers of drives, plus a comprehensive series of benchmarks intended to isolate the performance quirks of each RAID controller card at each RAID level. The results raise questions about whether IDE RAID can really take the place of a more expensive SCSI storage subsystem in workstation or small-scale server environments. Worthwhile reading for the curious sysadmin." I personally would love to hear any ide-raid stories that slashdotters might have.
IDE can only handle one or two hard drives per channel, which makes the cabling a real nasty hassle as opposed to SCSI-based RAID.
Even those so-called rounded cables can clutter the hell out of a tower case if you have a 4-channel RAID controller.
In my case it's the Adaptec 2400A four-channel, with four 120GB Western Digital hard drives, RAID 1+0.
Whats the point in having SCSI-Raid in most workstations these days? I mean, ram is so cheap now you can throw in a couple gigs for much less then the price diffrence between SCSI RAID and IDE raid.
I mean, I know the hest drives are SCSI flavor, but it seems like there's so many other things you could spend money on first that would get you way better performance, like getting a Dual Athlon CPU or something.
autopr0n is like, down and stuff.
their big old file server had 5 hard drives in it, but was only using 1 in windows! Being the smart boy that he is, he dutifully shuts down the machine, removes one of the drives, puts it on the broken machine, formats and loads windows on it.
So how did he decide which of the 5 drives he was going to pull ?
Oh. My. God.
:P]
I let out a yelp when I got to
puts it on the broken machine, formats and loads windows on it *
One of the things that really chaps my ass, more than anything else, is people asking my advice (and they do so specifically because of my experience in whichever field they're inquiring about), patiently listening to what I have to say, asking intelligent questions... then doing something completely or mostly against my recommendations.
More often than not, something ends up going wrong that would/could not have occurred had they followed my advice in the first place, and then I hear about it.
It sucks the last drop of willpower from my soul to hold myself back from saying "I told you so!" and charging them a stupidity fee. It's tempting to do so even to friends, if/when I get sucked into the resulting mess. [Hear that, Jared?
* Linux zealots: For a more warm-and-cozy feeling, disregard the first eight words of this quote.
Thats bullshit. Post some links to benches that back that up.
Two 80GB WD special edition drives in RAID 0 (7200RPM, 8mb cache) rarely burst over 90MB/s. They usually have a sustained transfer of ~50-65MB/s.
Additionally, your seek time is going to suck. I gaurantee its not going to be under 11ms. You cpu utilization during transfer will prolly be around 4% in the asolute best case senario and 11% on average. This is becuase, no matter what you think, all raid cards under ~140$ do the calculations for the transfers in software, not hardware. All you have is a controller card with special drivers. You wont come even close to beating the overall performance of a scsi 160 drive, or SCSI 160 RAID 0 setup.
Using IDE Raid is like using a winmodem. Unlike with modems, where everyone has one, RAID has a basic educational entry point. I seriously doubt IDE Raid will ever overtake SCSI in any area where knowledgeable people are doing the administration.
I have about 5 TBs of RAID5 storage online at various customer sites. They are all using Linux software RAID and Promise ATA66/100/133 controllers. Even when using two drives per IDE channel, we still see very good performance. An RAID5 system with eight 120-GB 5400-RPM Maxtor drives gives about 55 MB/sec write and 80 MB/sec read performance under Bonnie. Those eight drives were on two Promise ATA100 controllers. Cabling is fairly easy if you use 24" UltraATA cables. And it will get much easier with Serial ATA.
One customer ordered a system from a vendor who insisted on installing an ATA raid card, and it was a remarkable disappointment. Linux was able to indentify the array as a SCSI device and mount it. Then, for some reason, the customer rebooted his system. During the BIOS detection, the raid card started doing parity reconstruction and ran for over 24 hours before finally allowing the system to boot! For comparison, the same sized array would resync in the background under Linux in about 3 hours.
Also, the reconstruction tools built into the raid cards are pretty limited. If you have a problem with a Linux software RAID array, at least you can use the normal low level tools to access the drives and try to diagnose the problems. Just MO.
...so be alert.
Each IDE controller can support up to two drives, a master and a slave. What happens if you hang two drives off one controller, and the "master" drive dies?
If it dies badly enough, the "slave" drive can go offline. Now you've got TWO drives in your array that aren't talking. There goes your redundancy.
If your purpose in using RAID is to have a system that can continue operating after a single drive failure, then you better think again before you hang two drives off any one controller.
As it points out in the Linux software RAID docs, you should only have one drive per IDE controller if you're really concerned about uptime. That would imply that "4 channel" RAID cards should only be used with a maximum of two drives, both set to "master", and no "slaves".
Note that this does not apply to SATA drives, as there isn't really a master-slave relationship with SATA -- all drives have separate cables and controller circuits. SATA drives are enumerated the same way as older drives for backwards compatibility with drivers and other software, but they are otherwise independent. (At least that's what I hear, I haven't actually seen one of these beasts yet...)
And of course none of this touches on controller failures, which is another issue. But if you are worried about losing drives and still staying up, then better take this into consideration when you design your dream storage system.
(I don't know about you guys, but I have lost several drives over the years, and not one controller...)
I have a number of these stories to go along with my success stories. The problem is that at that point the technician is a saleman, your their technical advisor and the one who directly profits from their decision, this causes a certain amount of inherit distrust. No matter how well he explains it, he can't "force" the customer to do anything, it's their money. Only monopolistic corporations like say... microsoft (just a random pick) try to force their customers. That company was free to ignore the tech and put themselves out of buisness.
Here at work our main R&D server's been using a SCSI Mylex960 with RAID1 36GB drives. This has worked dandily for the past several years. This machine gets hit pretty had with tons of small IO, so I wouldn't consider IDE for it.
However, more recently we needed more builds/CDimage space so we picked up a Promise FastTrak100 (TX2) raid controller ($150CA) plus a couple 7200rpm 80GB Maxtors (~$150CA each), and have been living happily ever since. Now for sure we'd never put this in the main server, but for a cheap, reliable solution that gives you tons of space on a server that has only medium load, it can't be beat.
The point is, examine your needs and see what fits!
-Malloc
___________________ I want to be free()!
RAID 5 in software can be dangerous. If a parity write fails (disk/system dies), you'll likely have data corruption and not even know it. Best to trust reliable hardware to do the XORs.
Then again, a RAID _card_ may not help here, since the disks are at the mercy of the system power. Best to use a real array, if you have the bucks.
I think a new more accurate moderation is in order: Sad
-Ben
>What do you do if the controller card itself dies?
Simple... You purchase a different controller, put the drives on it, build the RAID, and restore the data onto it from your backups.
RAID is meant to increase overall reliability; it is not meant as a substitute for backups.
for my money its IDE raid all the way
---- Put Sig here:
I like reading the comments here, I am humble enough to know I can always learn something. But there's something I didn't see mentioned, in all these IDE RAID setups that people describe: can you have a hot spare disk? Hot spare is critical for data reliability. If you have a large RAID 5 or RAID 0+1 (not advised, always do 1+0, whenever possible), you can do the math and see how darn important it is to have the host spare.
What good it is to have a RAID 5 without a hot spare, when you can only guard against single drive failure? So, I really hope IDE RAID supports hot spare, otherwise I question the saity of mind of the admins who implement such solutions.
As for IDE vs SCSI drives, I have to say that I will always go with SCSI, as long as I am in a multuser environment where seek times are critical. Apparently (experience shows), if you put your database space on a RAID, seek times are critical for the performance of your application. In this context, I think this review/coparison would have benefitted from a real-life aplication's benchmarking, with a database hosted on the RAID.
Sigged!