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.
My experiences with IDE RAID have been pretty darn good. Benchmarking my Desktar 60GXP drives in Windows 2000 last year showed that I was getting read speeds in striping mode (between two drives) at faster rates than the fastest seagate Cheetah SCSI drives. Times have probably changed now though.
I started with a KT7A-RAID mobo. The important thing is that you get the cluster sizes just right for your particular partition. I used Norton Ghost to image my drive and try all sorts of different variables. In the end I had very satisfying results. Since I switched to Linux, I stopped using RAID-0 (yes, it is supported with this device!). I found that ReiserFS and the multi-drive Linux filesystem on these drives seemed to be just about as fast without having to hassle with soft-RAID controllers. It is probably due to my system RAM though. I couldn't seem to get Windows 2000 to make the most of 1024 MB without using that swapfile. Linux seems to avoid the swap altogether and uses static RAM instead. It is very nice having the extra IDE channels though. Without them, I probably wouldn't have 4 HDs hooked up right now.
I bought that about a while ago when the maxtor 160GB 5400RPM drives started to ship.
:) Reminds me when I plugued my first gigabyte drive in my amiga and saw big numbers :)
:).
I had to build a datacenter and storage price was the main issue. I had to have something cheap, yet hold a LOAD of data. Problem is personally I hate maxtor drives, I always found the more or less reliable (but drive experiences varies from a person to another so..). Anyways at that time maxtor were the only one offering 160GB drives, at a decent price/meg, and although 5400RPM is quite slow for access time, the main issue was cost so I could take a hit on access speed as long as "streaming" speed was fast enough.
the Adaptec 2400A card was the best at the time, simple, cheap efficient, it had 3 bad sides for my application, no 48Bits LBA support (130GB+), no 64bits PCI version (I was using a K7 thunder, and that chipset will slow down the pci bus to the slowest card connected to to bus, and since I wanted all available bandwidth to be thrown to the 64bits gigabit card, I couldn't accept using 32bits), and finally, no more than 4 drives. I wanted to break the terrabyte limit, so let's say I would have used 2 of those cards, it wouldn't have been price-performance-wise since the 2 would have shared the bus and I would have lost 2 drives for raid-5 instead of one with a 8 drive setup. but the performance of the Adaptec 2400A was the best. Still looks like the best overall today, yet I dunno if they are supporting 48bits LBA?
Anyways the 3ware 7850 was an excellent choice. Although their tech support is more or less good (like most tech supports) especially for real bugs and not just standard drivers reinstallation issues, the response time and sales people were very nice and professionnal. I got surprising results from the array, where I thought it would run like molasse, I was getting over 50MB/sec sustained non-sequential reading if I recall correctly. And the tools are very good, rebuild time is about 3-4 hours with 8x160GB @ 400GB filled on the drives, there are email alert tools and web interface to the host machine to check diagnostics. Overall it's a nice system and I'm sure the 7500 series are even better.
Oh and on a "funny" note, windows shows 1.1TB available in the explorer window, not 1134GB
As for the maxtor drives, I didn't take any chances, I ordered 10 to get 2 spares, 2 blew off in less than a month, but didn't have any problems since then, I guess if you can afford the time, doing a 1 month burn-in test with non critical data isn't overkill. usually they SHOULD blow up one by one so you could rebuild the array
--- Metamoderating abusive downgraders since my 300th post.
We've run several big RAID-5 setups on 3ware cards. When I say big I mean 1TB+ on each card. To do this we've used the 100GB+ drives available (120GB - 160GB) The biggest problem has been drive failures. Out of the 40 drives I think we've lost 6 in less than 1 year. In only 1 case have 2 drives gone bad at once (RAID-5, we're covered if 1 drive fails), but lost around 1TB of data. Luckily the data could be reproduced but took two weeks to regenerate.
It's WAY too easy to build massive arrays using these devices. How the hell are you supposed to back them up? You almost have to have 2, one live array and 1 hot spare array. If you think you're going to put 1TB on tape, forget about it. If you have the cash to buy tape technology with that capacity and the speed to be worthwhile, you should be buying SCSI disks and a SCSI RAID controller.
I have a dual Xeon 2.4GHz 4U with dual 8 channel IDE controllers connected to 16 160GB IDE drives under Windows 2000 arranged as two separate logical drives.
I'm able to read sequentially from very large files (20GByte+ files) at a continuous rate of over 180Mbytes/sec.
The controllers are 64-bit, 33MHz PCI cards and the high speed sequential reads are exactly what my application demands. SCSI would have added nothing to the performance of the system except an additional 60% to the cost.
Find me a 2.5TByte dual Xeon 4GByte RAM 4U box with SCSI drives for well under $10K and I'll give SCSI another look.
Once serial ATA comes out I think you'll see even more IDE based RAID being used.
Winmodems do the calculations through software because they lack the chips on the card. That's a horrible comparison. These ATA RAID cards have everything built on the card. The Promise SX6000 even has an on board Intel i960RM RISC processor for XOR calculations.
CPU utilization of these ATA RAID cards is negligible, so if you really need that extra 2 or 3 percent, just get a faster CPU.
The main advantages that SCSI has for performance is the individual drive performance (15,000 RPM and 4.5ms access time as opposed to 8.5) and command queueing. The transfer rate isn't a big issue if you're transferring it over the network. You're still limited to your PCI bus speed and the network speed. Even on a gigabit backbone, that's roughly 65MB per second of thoroughput in real world performance. The performance is only a factor for local reads/writes and access time.
The cost of a 1TB RAID 5 IDE setup (6 200GB drives, Promise SX6000 card, removable enclosures for the drives, and 128MB cache) = $2,450
The cost for a 1TB RAID 5 SCSI setup (8 10,000 RPM 146GB Cheetahs and an Adaptec 2200s dual channel card plus the hot swappable enclosures (add at least $700 here) = At least $9,350
If price is no object, go with SCSI. If you're running an enterprise SQL or WWW server with thousands of users, the access time of the drives is a huge benefit, so go SCSI. If each server must have more than 1TB of fault tolerant storage space, go SCSI because it can house enough drives per card to accomplish this. For everything else, go IDE.
As an FYI, I'm running the described ATA RAID 5 setup with 120GB WD Caviars with 8MB buffer, a dual port 3com teaming NIC, 512MB RAM, and an Athlon XP processor as a highly utilized file server. Runs like a champ. No issues and the boss is incredibly happy with the price tag. $2,800 to build the whole server. It's rackmounted under our incredibly expensive Compaq Proliant ML530 which is just doing SQL. If a drive goes out, I'll get an email notification. I simply remove the dead drive, replace it, and rebuild. No rebooting needed.
-Lucas