Slashdot Mirror
HOWTO: 0.5TB RAID on a Budget
Compu486 writes "Inventgeek.com has a new how-to article
titled 'The
Poor Mans Raid Array.' The article details how to make a modular .5 terabyte
Raid 5 array for under $250 (USD), and it all runs on the Mandriva flavor of Linux." Drive prices being what they are, this seems cooler than it is practical. Update: 06/25 23:31 GMT by T : If that's not enough storage, Yeechang Lee writes "Let me show off the 2.8TB Linux-powered RAID 5 array I built for home use a few months ago. I provide lots of details on how I did it, what I used, and the results. The Usenet thread has good followup posts from others, too."
That seems like a lot of screwing around.
Why not just hang a four *large* drives in a workstation with MB that does RAID 1+0? Yeah, it'll cost more than 249, but it won't involve a 50 lbs box of drives..
http://request-header.info
What would somebody with only $250 to spend on a .5tb raid have to fill it with?
I'd be somewhat more impressed if all of those drives he put in the system were in the 200-400gb range...
I know he got the drives on the cheap and all, but still... I've got nearly 1tb in my desktop machine without all the extra work.
N.
"Nothing strengthens authority so much as silence." - Charles de Gaulle
I too have this same exact problem and haven't been able to figure out what causes it. The array in question is a 1.6TB software RAID5 array with eight 250GB Maxtor SATA drives on Promise SATA150 TX4 controllers. A few months ago I noticed files that I'd coppied via samba being corrupted once they got to the array. As such I now checksum every file that I dump on it before copying from my windows box. I would say somwhere between 1/5 and 1/10 of the large (350MB+) files I copy to it get corrupted. If I copy them again, the MD5s are different, and usually the second copy gives me the same checksum.
Things I've noticed:
- swapped out switches, network cables, and ethernet cards with no change
- occurs going from my Windows box to the RAID array with Samba, generic FTP, and even scp
- only seems to occur when copying to the RAID array, not when copying to the non-RAID system disk on the same server
- happened with both an old Abit BE6 and with a newer Intel L440GX motherboard - I don't think the chipsets have anything to do with it
- pretty sure it only happens between my windows box and the linux server, never from another box (which includes Linux, FreeBSD, and OSX)
- only happens when I copy to the RAID array, reads seem to work flawlessly
- strangely this only seems to occur with
.avi files - I can't recall having any other types of files become corrupted but as a disclaimer 95% of the things being coppied to this server are AVIs
I've no idea what causes this, and as far as I know it only started happening somewhat recently (I've had the array for many years, although not always a full 1.6TB in size). Maybe it started when SP2 came out, I don't know. I can't 100% confirm I haven't always had this problem, but I've only noticed it starting a matter of months ago. Anyone else out there have any ideas? I'm at a loss. The Promise TX4s are what I'm currently eyeing as the problem source.FYI: I'm using mdadm 1.11.0, ReiserFS 3.6.19, and a chunk size of 128k with left-symmetric parity.
I spend my entire life managing large SANs, so RAID is done in the array (EMC, HDS) while basic volume management is done on the host (LVM, VXVM)... so when i first read this I thought that somebody had used linux and a fibrechannel HBA running in target mode (http://www.emulex.com/ts/docfc/linux/430l/target_ mode_intro.htm)
/. and you'll have something b/c you'll have shown something more than 'look what linux can do' that the other OS's have had for years...
Put that up on
And then going on to mount those luns on another system (say a solaris, aix or another linux box). Instead, I was dissapointed to find out that you took a linux box and created enough software RAID to for a TB or more. If this was done with windows, it would be rejected... so why doing it with Linux make it front page news?
As mentioned, my 2.8TB setup uses LVM2 on RAID 5 (mdadm, not raidtools). I think anyone building one of these babies would be crazy to not use LVM; why limit your future expansion options?
I recommend creating one primary partition which is slightly smaller than the fullsize of the harddisk
I've built a ~600GB RAID array for my home video jukebox, and I'd modify your recipe in one way: Rather than creating just two partitions on each drive, create many, then create many small RAID arrays and glue them together with LVM. The result is much more flexible. You can use different partition sets with different RAID levels for different purposes, and it also makes adding additional storage into the array very easy.
In my case, I have four 200GB drives, so I created 10 20GB partitions on each. Eight of the 10 partitions on each drive are used in RAID-5 arrays, one is used in a RAID-1 array (mirroring, across all four drives), and one is used in a RAID-0 (striped) array.
This initial configuration gives me 80GB of very fast RAID-0 "scratch" space, for storage of data that is not important, 20GB of highly-redundant RAID-1 space for data that is very important, and 480GB of moderately-redundant RAID-5 space for everything else. I set up three volume groups called 0g, 1g, and 5g, containing the appropriate RAID arrays, then carved out logical volumes as needed.
But, as my needs change, I can shift those RAID sets around. Supposing, for example, that I decided I needed more RAID-1 space, I could pull one RAID-5 array out of the 5g volume group, rebuild it as a RAID-1 array, add it to the RAID-1 volume group and then allocate the storage to whichever logical volume needs it.
Not only that, adding new storage becomes easier. With only one large RAID array, adding a new disk to the array requires backing up the entire array, rebuilding it with the additional disk and then restoring the data.
With many small arrays, as long as there's enough free space that you can remove an array from the volume group, you can rebuild it one array at a time. I actually started with three 200GB disks, so each of my RAID-5 arrays held 40GB, not 60GB. When I added the fourth disk, I wrote a small script that went through the eight RAID-5 arrays, pulling each one out, destroying it, rebuilding it with four disks and reinserting it into the volume group. The whole process took about a day to run, but my data stayed both available and safe the whole time, and didn't require me to figure out where I could back up nearly 200GB worth of data.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
1) Free -- Machine was an old K6-2 500 (192MB RAM, 1.4G Boot Drive) that I had laying around.
2) Free -- I got a full tower case from my brother in law (no faceplate).
3) Free -- I had a few 120mm fans laying around which I have cooling the drives.
4) $1040 -- 8 Maxtor 250 GB PATA HDs. (8MB cache, 7200 RPM)
5) $215 -- 3Ware 7810 (8 port PATA hardware RAID 5 card).
6) $140 -- APC RS 1500 battery backup. (You don't want the array to suddenly lose power for any reason!)
Total Cost $1395.
What it got me: I have 1400 GB usable redundant storage with a hot-spare. If a drive fails at 1:00am the computer will automatically start the rebuild on the spare drive, and likewise if I'm not home. This was more important than the additional storage. I also know that I can get 40 minutes of power out of the APC if the power goes out. The machine is set up to shut itself down in the event that the battery runs low.
I didn't have to fight with any software configs. The driver is included in the Linux kernel source, and can be compiled into the kernel. I don't have to worry about figuring out SMART data. "tw-cli info c0" gives me easily readable output on all of the drives plugged into the RAID card. It's simple, does the job, is stable as all hell, and was fairly cheap. It would have cost nearly as much to have bought 4 PATA cards (ones not using the flawed silicon image controller) as it cost for the 3ware card off of eBay.
More information here.