SATA RAID Enclosure w/ Temperature Monitoring?

vanyel asks: "Yesterday, my external USB 2.0 drive enclosure finished cooking a 3/4 full 200G drive after its fan quit working who knows how long ago. In the time honored tradition of closing the barn door *after* the horse has wandered away, I'm accelerating my quest for a RAID solution. In particular, I want something that will support 4 SATA drives and has temperature monitoring that doesn't require a particular vendor's RAID card or Windows. Better yet, is there a RAID-5 NAS that isn't in the $4-5000USD price range. Anyone with a better barn door to close this problem with?"

Comments

Use SMART?

[crow]

At least with ATA drives, you can usually use smartd to monitor your drives. This includes temperature and various failure indicators. Usually when a drive fails, there is plenty of warning from small failures that the drive recovers from. When you run smartd, you can receive these warnings.

Re:Use SMART?

[Sepper]

I second that. You can have Smart Run a script when things fails, like Mailing the details of the failiure and then turning off the computer.

For other computer parts lsensor might do the trick.
For exemple: http://gentoo-wiki.com/HOWTO_Monitor_your_hard_dis k(s)_with_smartmontools

if it's an external device, the best thing would be to get a controlable UPS And turn off (again with a small script)

Just think RAID, UPS, smart monitors and deamons and with a bit of imagination, you can come up with a soultion.

Re:Use SMART?

[Old Uncle Bill]

Yeah, and don't rely on Windows smart detection. After my 99% full 250 GB drive failed last week, I was able to resurrect some of the data. Included in this was the event logs, with about 500 warnings over two weeks that my drive was on the way out. Okay, so in order to figure out if my drive is going south I need to check the event log every day? WTF?

Re:Use SMART?

[isorox]

Every week would have done. Simply have a daily cron job that greps you logs for fatal messages and email them to you.

Re:Use SMART?

[endx7]

And best of all, SMART works with not just ATA, but SATA too, which is what asker seems to be asking for.

Re:Use SMART?

[Anonymous Coward]

Read your post,OVER TWO WEEKS..

I bet if you looked say, every monday, you might have caught it.

Re:Use SMART?

[hackstraw]

Okay, so in order to figure out if my drive is going south I need to check the event log every day? WTF?

Yes you do. Its no big deal at least for operating systems that log in plaintext (I don't know about windows). What I do is nightly grep for unusual stuff and I take 30seconds to a minute reviewing it for all of my systems and I admin about a hundred, so one machine should take about 5 seconds or so a day. That is much easier than trying to recreate 250 Gigs of data from a dead drive. Right?

Re:Use SMART?

[raxhonp]

And then use hddtemp to ease monitoring your drives.

Re:Use SMART?

[Anonymous Coward]

I wish someone would slap this into the ports system of FreeBSD.

Re:Use SMART?

[richie2000]

Okay, so in order to figure out if my drive is going south I need to check the event log every day?

When I figured that one out the hard way was also the exact moment I decided to convert from Windows on the servers to Linux. Real servers log plain-text.

Re:Use SMART?

[paganizer]

Windows does log to plain text.
Finally, a /.'r admitting that Win2k server is a "real server"... and wait, it's been about 0 degree's F here in N. Central KY (otherwise known as hell) for the last few days....
Ragnarok is coming!

Re:Use SMART?

[richie2000]

Windows does log to plain text.

Really? Do you claim C:\WINNT\security\edb.log is plain text just because it doesn't contain any Unicode characters? C:\WINNT\system32\LogFiles would be a nice place for the logs, but it only contains the IIS logfiles. C:\WINNT\Debug has a few other logfiles, but nothing consistent and no system.log like you'll see in the Event Viewer.

OK, I'll end the suspense. The system log is at C:\WINNT\system32\config\SysEvent.Evt . First place to look for a log file that is, the config directory. Yep. Sure.

BTW, I dare you to read that logfile in Notepad. It ain't plain-text.

Re:Use SMART?

[paganizer]

I just read the file in notepad.
Next bet?
(um, it's not EASY to read it in notepad, thats not the tool I would use by choice, as it's not formatted to be read in notepad).
I used to keep tabs on a 70 sub-office WAN network by occasional checking the event logs.

Re:Use SMART?

[richie2000]

You must have a weird eye-brain parser.

Now, let's see what Slashcode does to this snippet from my logfile (actually, one of my customer's):

†晌敌 † †&# 49324; 쁄 四 &#19585 ; † ​†㪀 & #137;†††Ԡ‒†   ⁐⁒⁉⁎⁔⁓& #8261;⁒⁖⁅⁒ ⁐&# 8241; ⁌⁊⁕⁓⁄&#8 257;⁌ ⁓⁥⁳⁳&#82 97;†‷†††&# 128;†€†晌敌嬋&#81 94;仹䇈仹䇈 耠   ††††⁞&#134 ;‌†⁒†††₤&#13 4;⁐⁲⁩⁴†⁓ ⁅⁒⁖⁅⁒†ā&#1 34;†Ԡ‒†  ⁐ ⁒⁉⁎⁔⁓⁅⁒& #8278;⁅⁒ ⁐‱ &# 8268;⁊⁕⁓⁄⁁⁌&#8 239;⁓⁥⁳⁳⁩&#83 02;†‷†††€†&#12 8;†晌敌嬌 仹&#1 6840;仹䇈 耠   ††††⁞†‌† ⁒†††₤†⁐&#830 6;⁩⁴†⁓⁅⁒ ⁖⁅⁒†ā††&#131 2;‒†  ⁐⁒⁉ ⁎⁔⁓⁅⁒⁖⁅& #8274; ⁐′ ⁌⁊&# 8277;⁓⁄⁁⁌ ⁓&#8 293;⁳⁳⁩†&#824 7;†††€†€†&#2618 8;敌嬍 仹䇈仹& #16840; 耠  ††&# 134;†⁞†‌†⁒†& #134;†₤†⁐⁲⁩&#830 2;⁴†⁓⁅⁒⁖⁅ ⁒†ā††Ԡ‒&#13

Re:Use SMART?

[richie2000]

OK, let's try it raw (the previous was saved in notepad, read back in because it contained characters that Slash couldn't handle. This little bit seems to work. And no, I don't call this plain-text, even though parts of it is plain enough. No datestamps, nothing except parts of messages. Is there a "Create legible logfiles" option in the registry that you have found and turned on?

0 LfLe À DÀ ÛV L : 0 _ _ P R I N T S E R V E R _ P 1 / L J U S D A L / S e s s i o n 7 LfLe[ ùNÈAùNÈA LfLe0[ x"ÉAx"ÉA Ô Ô LfLe+[ û!ÉAû!ÉA ^ R Ì

Re:Use SMART?

[paganizer]

Odd, in mine it comes out like this:

E v e n t L o g S H A D O W 1 4 : 3 6 : 0 4 P M 1 2 / 1 4 / 2 0 0 4 Ô $ mÔ $ m LfLe bAbAy

Maybe you aren't opening the right file? I had to take out enough to get by the junk filter, but aside from some non-standard characters and spacing between letters, it's completely readable.
it's SysEvent.Evt; there are other system files there, and if you have your file extensions turned off (shudder) it's easy to click on the wrong one.

Re:Use SMART?

[richie2000]

Nope, it's the right file. In fact, none of the files in that directory seem to be readable. Maybe it actually IS a registry setting somewhere. :-)

aside from some non-standard characters and spacing between letters, it's completely readable.

So, can you tell me what is actually says in your example, apart from the datestamp? What does a disk read error look like, for example? That's the one I would have wanted to know about before going home for the weekend while the server happily thrashed a few hundred gigs of pr0n.

What's worked for me is...

[shadwwulf]

the nStor 4700 SATA series. See http://www.nstor.com for more info.

MTW

Re:What's worked for me is...

[Mitsoid]

nStor 4700: ~$4-6,000, depending on specific model.. no idea how many drives it comes with.. at least, that's the only price I could find after looking arount the web for the last 20 minutes (since the parent thread first was posted) I think the Topic was looking for a NON-$3,000 to $5,000 solution... which I must agree, I am too... 4 SATA 250GB HDD's costs less than $1,000.. a RAID 'card' & case shouldn't cost $4,000 for non-business use.. I have over 500GB in work already, but I can't really back up most of it (I try to compress and burn to CD a lot--most of the space is my TIFF/RAW digital photos, which compress well, but I edit them too much to leave them as such).. I'd rather spend $1,500 on an SATA RAID solution then on a tape drive & tapes...

Re:What's worked for me is...

[Kosi]

Yes, I searched for a price in vain for 10 minutes, too. But I found many articles saying something about the nStor 4700 being really cheap, so one could assume that $4-6k is the price complete with 12 400 GB drives.

What the market completely misses is a easy and comfortable storage solution for the SOHO or advanced home user at a reasonable price. Not those stupidly small 1 or 2 drive boxes that cost even more than a complete PC including a drive of that size, and not those absolutely insane-priced "professional" ones.

Just an appropriate case with 4, 8, 12 or 16 bays for S-ATA drives and some embedded system including the RAID 5 controller and a Gbit Ethernet port. This should be a lot cheaper than rolling your own with a complete PC, which would cost 300 to 400 Euro for the 8 bay version.

freebsd and vinum (to avoid the card)

[TheGratefulNet]

I just recently setup a vinum test for freebsd. I ganged 4 200gig drives and got this:

# df -h
Filesystem________Size___Use___Avail__Capacity __Mounted on
...
/dev/vinum/raid___734G___578G____97G____8 6%____/vinum

(sorry for lameness in formatting. LIT doesn't work on slash...)

(yes, 4*200=800 yet its really only 734)

anyway, what's neat is that I used a variety of controllers and it all worked. I first started off with a $15 ide controller and 4 ports (2 channels m/s). used SiL chipset. worked fine. then I moved one drive to sata and used a bridge to convert sata to pata (the drives I have are regular IDE parallel). that worked too!

then I moved them to the motherboard master/slave and that worked with the same config as well (just update a single /etc/ file and restart vinum).

so all that gets you raid in software and - FREEDOM to choose a diff controller (even sata or parallel) and it will still work.

can't help with the temp controller though.

might make sense to use drive diversity, though. and mirroring. so if one drive 'cooks', likely the other brand won't. and with software raid, drives can be of diff makes and models and it will still work ok.

NOTE: vinum works well on bsd stable (4.x). do NOT use it on 5.3 - its not really ready yet (I gave up and went back to 4.10).

Re:freebsd and vinum (to avoid the card)

[drakaan]

In a RAID setup where you're hoping to have some kind of redundancy, shouldn't 4 x 200 equal 600 or 400?

losing a single 200GB disk in that setup means that you most likely say goodbye to 734GB of data, rather than giving said disk a proper burial and replacing it.

Re:freebsd and vinum (to avoid the card)

[TheGratefulNet]

I use true hardware mirroring for the o/s drive. 3ware /dev/twed0 is great.

for my data pack, I use concat (linear) mode. its not striping since it doesn't paint across all disks for each next block. it fills up one disk at a time, then moves on to the next. if I stay within that disk, no others need be accessed. this allows the mobo to spin the others down and save heat, power and life of the drive.

but what if one of those fails? it will probably bring the whole pack down, like you say.

in that case, you must have backups. I do. my friends are my backups. my large data back is mostly audio+video. its too expensive to have 2x the storage for backups of this stuff, so I give copies to friends and we are a mutual backup of each other. and we get to enjoy the content, so its not wasted 'shelf spares' that sit around and do nothing, most of the time.

but you're right - you MUST know the risk of data loss before you choose any storage technique. my system's uptime was very important, so I used 3ware hardware mirroring on 2 quiet seagate 7200.7 drives. for my data pack, that's a whole other matter - and my life doesn't stop (so to speak) if that pack fails. but my system pack MUST be up always and never have data loss or corruption.

choose the right method for the right task - even if hybrid (like mine).

Re:freebsd and vinum (to avoid the card)

[TheGratefulNet]

(sorry for the quick followup)

forgot to mention: the main reason I went with software raid for the data-pack is that its GROWABLE non-destructively. I asked 3ware and some others (highpoint, promise) if their controllers supported how disk-adds to add more storage to a pack. none had it. that really surprised me.

I didn't want to be a slave to any one vendor for my almost-terabyte media collection. if I need to grow this array (and I will, periodically) then freebsd vinum on 4.10-stable seemed to be the very best choice (assuming pc hardware, etc etc).

hardware mirror for the system pack. and anything that you need growable, I really strongly suggest you consider a software implementation.

Re:freebsd and vinum (to avoid the card)

[Fweeky]

Did you try gvinum? I gather the original vinum's been getting rusty in 5, and has been removed entirely in favour of gvinum in HEAD at least.

gmirror and gstripe are also well worth looking at.

Re:freebsd and vinum (to avoid the card)

[drakaan]

Interesting...you get none of the redundancy or preformance benefits of, say RAID-5, but growability seems to be your only real-time concern.

You might consider RAID-5 for the physical disks in the data pack with concat on top of that (if it's possible). Then you have hot-swappable redundant data storage that you can grow as you please (albeit 3+ disks at a time). Still hybrid, but trading some space for better availability. Done this way (if it's even possible), dropping a physical disk won't kill your data pack.

Re:freebsd and vinum (to avoid the card)

[TheGratefulNet]

right, I could layer this logical topology over some other abstract phys topo. raid 5 or any other.

for storage that DOES grow (my media/music collection) growability is prime. rebuilding is NOT an option just to add more logical space.

it seemed that going vinum (or gvinum for bsd5) is the right logical layering.

and its nice to isolate dealing with failed spindles and dealing with growing space. my space problem is solved by vinum. the spindle problem is solved by having backups via friends. no, its nowhere near realtime to replace, but I save on moving parts and noise and heat. and in a typical home environ, yes, that matters.

my enclosures do not have fans...

[humuhumunukunukuapu']

both are 7200 rpm drives and are about a year old...should i worry?

Re:my enclosures do not have fans...

[TheGratefulNet]

for IMPORTANT stuff, I buy a new drive each year.

not only is it faster (and lately, quieter) but you don't worry about spindle locking (seizing) or crashes - but you also can rest easier.

rule: if you can re-download it, don't worry.
if you write it yourself, BACK IT UP on multiple brands/types of drives.

Re:my enclosures do not have fans...

[harrkev]

for IMPORTANT stuff, I buy a new drive each year.

And for an additional $30, you can turn the old drives into a portable USB drive. Each month, grab one and copy all of your important stuff to it. Then, tote it over to a friend's house for safekeeping. Then, even if your house burns down, you still have all of your old data, not more than a month out of date, for only an additional $30.

For best results, use two separate USB drives, and do not have both in your house at the same time. It would suck if a trojan or virus wiped your HD, and your backup USB HD that happened to be backing up at the time you got hit.

Re:my enclosures do not have fans...

[TheGratefulNet]

while you're at it, get a proper usb2 AND firewire enclosure.

fw still seems more robust and reliable. not sure why, but it is.

oxford 911 chipset is your friend.

then again, serial-ata is becoming an EXTERNAL standard and so that's even better than fw or usb2.

3ware

[Bad Boy Marty]

I've used 3ware's cards in hard-core production machines for years, and I just can't say enough nice things about them. Even bashing them with bonnie (disk throughput benchmark), they give better performance than Dell's PERC SCSI/RAID controllers. Now that I've had a need to use them at home, I'm still ecstatic with them. Smartd has had the capability to monitor drives on these controllers for a *long* time, and 3ware's own monitoring software (although a little clunky in a couple places) offers a bunch more than smartd.

If you want NAS, just throw the controller & drives into a Linux box, and export the filesystems through Samba, and all the machines on your network will be happy to connect.

No, I am not related to, nor paid by anyone related to 3ware. Just a *very* happy customer.

Re:3ware

[TheGratefulNet]

NOTE- I have 3ware cards (7000 and 8000 series) on my freebsd box (/dev/twed0).

works great - BUT - one big big catch. no online extends! ;(

I had 2 drives in concat raid (striped) and wanted to add more storage. NO WAY TO DO IT unless you tear it down, break it down and totally rebuild. which means I have to move that data OFF the pack and then reformat the new pack (new drives added) then copy back.

sucks!!!

for a hardware card that costs a lot - its sucks in terms of online extends.

for mirroring, its great. that's what I use it for. all my bsd boxen have twed cards in them. but for striped large data sets, 3ware sucks.

I mirror my system pack in hardware (vinum is tricky in software as a system boot disk) but for 'data disks' I use software raid since its MORE FLEXIBLE and doesn't show limits.

oh, also - with motherboard ide controllers you can set the auto power down (spin down) to as low as a minute, and then the drives will spin down (no heat or noise) if they don't get accessed in timeout. works perfectly! in both linux and freebsd. if a drive is spun down and you (the user) want to access it - it will auto spin up (seamless to the o/s) and when its spun up again, the os accesses it normally). you just see a slight startup delay that's all.

I do that for my mp3 collection since its not always in use but its always online for me. when I play a song from my raid setup, I see a few seconds delay as the drive spins up, then I hear the music play. no big deal - and I like knowing the drives aren't spinning 7x24 for no good reason.

no hardware controller I know of (I've used a lot..) does this. it MUST be in software and usually on the motherboard built-in ide ports.
if you want the auto-power down, that is.

Re:3ware

[compwizrd]

9500S supports OCE.

Re:3ware

[TheGratefulNet]

I called tech support just a few weeks ago, in fact, asking if any of their controllers, now or in the near future, will support on demand expansion.

they said no, but it is being worked on.

JUST being worked on NOW?? wow...

my ancient mylex card (long format pci, scsi) could extend, I'm pretty sure.

but things are not as advanced in ide-raid, it appears. even from the 'big guys'.

Buy a heat alarm for $10

[Anonymous Coward]

PCPowerCooling.com sells an overheating alarm for $10. I put it in all the systems I build.

Alarm Available Here

Re:Buy a heat alarm for $10

[stienman]

According to the ad copy it'll even work in your 486DX system!

I just found it funny to see the picture demonstrating the device in a computer that needs no special CPU cooling.

-Adam

Re:Buy a heat alarm for $10

[TheGratefulNet]

if you need measurable readings, you could always buy a DVM that has temperature probe ability - and a serial port (a lot of voltmeters have that these days) and then talk to the DVM via the serial port and poll it for temp.

not all that hard, really. I bet a dvm to do that might be $50 or less.

if you have an older mobo (old asus boards, before the built-in temp probes for cpus) had these 2 pin headers where you could plug in a thermister and fit it near the cpu to get the reading. that reading could be read via i2c (or is it i2s?) the sysmgt bus that linux has supported for years, now.

just wire that to a long wire that goes near your drives.

but all modern mobos have this already built in, or the cpu itself has it. you'd have to look at least 3 yrs ago to find one that takes an external temp probe via the 2pin header.

Re:Buy a heat alarm for $10

[Cheeze]

i dunno, i left my 486sx33 by the window once and the sun got it. Nothing was damaged, but it wouldn't boot up for hours, until it cooled down.

The original poster should just spend the $12k and get an Apple XServe.

Avoid RAID5

[photon317]

There is really only one good reason to ever use RAID5, and that is that you're too tight on money to be able to afford to RAID1 (Mirror) the storage you need (If you need 400G of space, RAID1 is gonna cost you 800G of storage, whereas RAID5 might only cost you 500G of storage). RAID1 is both faster (For writes and especially reads) and more resilient than RAID5. Assuming you can afford it (and storage itself is pretty cheap today, especially if you don't get a fancy RAID5 controller), just go with RAID1.

If you want really nice performance and you're buying 4+ drives, do RAID1+0 - mirror the drives up in pairs (where the pairs are as diverse as your setup allows, seperate controllers and/or chassis and/or power, etc...), then stripe the data volume on top of the sets of mirror-pairs.

Re:Avoid RAID5

[TheGratefulNet]

raid 5 isn't fast, usually.

but sometimes you HAVE little choice. if you have massive storage, you can't very easily full mirror it.

what if your case is already filled with drives? what if your power supply can't handle 2x the drives?

raid 5 is fine. slower, but not a bad setup if your controller does xor in hw.

Re:Avoid RAID5

[Anonymous Coward]

raid 5 is fine. slower, but not a bad setup if your controller does xor in hw.

Until you have two drives fail. Then, you're fucked. Don't act like that never happens, because it does. I've had it happen, as I'm sure others have. No more RAID5 for me. . .

Re:Avoid RAID5

[TheGratefulNet]

true.

no array is ever completely fault tolerant.

you STILL need backups.

but raid helps get you buy during the 3am disk failure and you don't want to drive 50miles to replace a failed disk.

in the AM, when you get to work, THEN you replace it.

raid is not subs. for backups. but it helps get you thru the single spindle failures.

its better than NOT having it.

Re:Avoid RAID5

[itwerx]

PP - raid 5 isn't fast, usually.
Depends on the controller. If it supports parallel reads/writes it can be the fastest of all RAID configurations. (Otherwise it's the slowest :).

P - Until you have two drives fail.
Well duh! That's true of any RAID solution. And while it does happen occasionally, I can say that it's pretty goddamn rare! I've been consulting for over ten years and out of many thousands of RAID configurations across numerous controllers/drives/etc I have seen exactly one case of honest-to-goodness simultaneous drive failure.
Now controller failures that trash the RAID are another matter entirely - seen that lots of times! :)
As with anything it just depends on the quality of the gear...

Re:Avoid RAID5

[secret_squirrel_99]

Until you have two drives fail. Then, you're fucked. Don't act like that never happens, because it does. I've had it happen, as I'm sure others have. No more RAID5 for me. . .

Thats what hot spares are for. Even if you aren't monitoring your arrays.. you are aren't you? One global hot spare per enclosure and you need 3 failed drives before you loose data.

Re:Avoid RAID5

[hackstraw]

no array is ever completely fault tolerant.

you STILL need backups.

I'm not sure what the target use is, but it seems like its personal, and being that the previous external drive was only USB, performance does not seem to be a concern.

With that in mind, I would suggest poor man's RAID1 over real RAID1. By that I mean buy two disks and cron a rsync command every night. This would take care of backups and redundancy, although its not realtime, so a disk failure after a disk write but before the rsync would loose your data, but that is very unlikely and if something is that important, extra precautions should be made. In my opinion, real RAIDx is only necessary if uptime is of importance. For personal use, I would guess that a quick switching of disks or mounting another one and making a symlink is OK.

In other words, I agree with the theory of RAID1, but RAID1 + a backup requires at least 3x the storage space. I was looking at getting two Lacie bigdisks (500Gigs) and do this mirroring on them, but after the one died at work, and a quick google search says that most all of the Lacie RAID0 disks rarely survive more than a year, I too am in the market for an enclosure. In looking at the Lacie enclosure, it doesn't take someone too long to figure out why they all fail. The airflow is from front to back and one harddrive is directly in back of the other, so the 2nd drive gets all the heat. Doh!

I would really like for something like the Lacie bigdisk that was properly engineered. I would be comfortable with having about 500Gigs of space for a while.

Re:Avoid RAID5

[Fweeky]

"P - Until you have two drives fail.
Well duh! That's true of any RAID solution."

Really? I kinda hope that if I have two drives fail on my 3 drive RAID-1, the array's not going to mysteriously disappear.. and I'd hope my RAID-10's can survive at least some combinations of multiple drive failure :)

Re:Avoid RAID5

[itwerx]

Me - "Well duh! That's true of any RAID solution."
Fweeky - "Really?"

True enough, maybe I should have been a little less emphatic, eh? :)

Out of curiousity, (not argument), since there are no "standard" RAID definitions above Raid-5, which particular manufacturer's RAID-10 were you referring to?

Re:Avoid RAID5

[drsmithy]

raid 5 is fine. slower, but not a bad setup if your controller does xor in hw.

The RAID5 slowdown has nothing to do with parity calculations (not these days, anyway).

Re:Avoid RAID5

[drsmithy]

There is really only one good reason to ever use RAID5, and that is that you're too tight on money to be able to afford to RAID1 (Mirror) the storage you need (If you need 400G of space, RAID1 is gonna cost you 800G of storage, whereas RAID5 might only cost you 500G of storage).

Well, if you're going to be accessing the data in such a way that RAID5's dismal write performance isn't an issue (eg: 90% reads, over a 100Mb network, etc) then RAID5 is by _far_ the better bang/$ solution.

Re:Avoid RAID5

[drsmithy]

Depends on the controller. If it supports parallel reads/writes it can be the fastest of all RAID configurations. (Otherwise it's the slowest :).

RAID5 is always going to be slowest of the "common" RAIDs (0,1,5) for disk writes (discounting seriously broken hardware and/or software) simply by virtue of the way it works.

RAID5 reads should be as fast as anything else.

Re:Avoid RAID5

[Fweeky]

Hmm? How is layering RAID-0 on top of RAID-1 not "standard"? We're using Adaptec 2210S' if that makes any difference.. if it's not layering basic RAID levels like it says it is wtf is it doing? ;)

Re:Avoid RAID5

[Anonymous Coward]

Stop, you're scaring me! I have two of the 500GB Lacie Big Disks. I do not use them heavily at all, they are more for deep storage and occasional access, so hopefully they will hold up fairly well...

Re:Avoid RAID5

[illtud]

RAID1 is both faster (For writes and especially reads) and more resilient than RAID5.

If you've a proper controller and a sensible number of drives, RAID 5 is by far the fastest in reading. Try getting a sustained 400MB/s out of RAID 1 and get back to me.

Re:Avoid RAID5

[Don'tTreadOnMe]

There is a second good reason: You need multiple terabytes of storage and don'w want to have double that many drives lying around.

First off,

[bplipschitz]

depending upon your data situation, how about RAID 1+0 or RAID 0+1 rather than RAID 5? RAID 5 will bite you in the ass, sooner or later.

I've got the teeth marks to prove it.

Re:First off,

[Hast]

ALWAYS put RAID 1 on the lower level and then use RAID 0 to bunch these two devices together. Reason being if you do that your array has a 2/3 chance of surviving two HDD crashes at the same time.

For a lot of data this seems a bit redundant IMHO. There is a difference between data you can recreate but don't want to loose (ripped CDs etc) and data you can't recreate (documents, photos). Depending on your storage needs it may be redundant to put it all on the same type of array.

Re:First off,

[TheLink]

Uh I don't see the difference in redundancy and performance.

Between

AB (AB)' (striped first)
which is the same as
AB A'B' (striped first)

vs
A A' B B' (mirrored first)

(' = mirror)

Whether you have RAID0 first or RAID1 first. As long as two As or two Bs don't fail you're OK.

In practice I suppose it's better to have the mirrors in physically different bay areas and attached to different controllers (some RAID controllers seem to have a nasty habit of failing before the drives fail).

Re:First off,

[Hast]

I find your notation a bit hard to grasp. I tried to invent my own but it seems like drawing pictures of it just makes it harder than just text. (I probably just haven't been clever enough to think of a picture though.)

If you first stripe and then mirror and one drive fails you have one partially failed subarray and one fully working subarray. If one of the drives in the fully working subarray fail the entire array fails as that subarray has the only working copy of the array. If the still working drive in the partially failed array fails then the array continues to work. In other words, there is a 1/3 chance that the "right" drive fails.

If you first mirror and then stripe the system and one drive fails you have two functional subarrays. Now if the lone drive in the damaged subarray fail the entire array fail. But if one of the disks in the fully operational subarray fails the array continues to work. Iow, there is a 2/3 chance that one of the "right" disks fail.

Well that's how I reasoned about it. Please feel free to point out any mistakes. I haven't considered any points about performance issues.

Re:First off,

[drsmithy]

Uh I don't see the difference in redundancy and performance.

If you stripe then mirror, then losing a single drive makes your entire array non-redundant, effectively turning it into a RAID0 (and greatly increases your rebuild time when you replace the failed disk as half of the entire array needs to be rebuilt). To lose the entire array in this scenario you only need to lose a single disk from the other side of the mirror.

OTOH, if you mirror then stripe, then losing a single disk only affects a fraction of one of the mirrors (and your rebuild time is shorter since only one drive needs to be rebuilt). To lose the entire array the next failing disk has to correspond exactly to the other side of the mirror of the already-failed disk.

IOW, it's the probability of a single disk anywhere on one side of the array failing vs the probability of a specific disk on one side of the array failing.

Of course, this only really becomes an issue worth seriously considering once you get into arrays with dozens of disks ;). However, in the interests of good practice you should still RAID1+0 rather than RAID0+1 (ie: mirror then stripe rather than stripe then mirror). If you do the maths, though, you should see that RAID10 is "more redundant".

ultra-cheap solution

[Anonymous Coward]

If your drive enclosure is not locked away in a cabinet somewhere, why not grab one of those cheap car dash thermometers and attach it to the outflow of the fans?

Supermicro has something that'll work

[the_maddman]

Supermicro makes the SATA drive cages I use, they have an alarm on them if the fan quit, or if they overheat, and it's loud enough you'll do something about it just to shut it up. Take a look at http://www.supermicro.com/products/accessories/mob ilerack/CSE-M35S.cfm

I got mine from http://www.newegg.com/ for around $150 when you get shipping and tax involved, and they work good.

roll your own!

[Kosi]

Get a good sized case with enough slots for fans, a S-ATA RAID Controller (from the Highpoint 1820 at ~ 180 Euro to the 3ware Escalade 9500 at ~ 550 Euro) and a decent fan/temp controller which is supported by your OS or ships with drivers for your OS.

Complete this with board, cpu and all the rest fitting your needs and you'll have the most S-ATA fileserver you'll get for your money! :-)

I just do the same for me, except the fan/temp controller.

Call me Mr Pedant, but...

[Singletoned]

Isn't the saying 'close the stable door after the horse has bolted'?

Horses live in stables, not barns AFAIK, so it would make more sense.

Re:Call me Mr Pedant, but...

[Tower]

Eh, it is usually the barn door, but it is also the cows, not the horses...

Re:Call me Mr Pedant, but...

[Singletoned]

" Eh, it is usually the barn door, but it is also the cows, not the horses..."

So, 'close the barn door when the cows have bolted'? Must be an American thing...

Re:Call me Mr Pedant, but...

[deque_alpha]

Yeah, but see, stables are in barns and therefore it doesn't matter if they get out of their stables, but if they get out of the barn... now that's trouble....

Many fans are fixable

[Daniel Rutter]

Most computer fans have the decency to make a racket for a while before their bearings are so damaged that the fan stops, or doesn't spin fast enough to keep its component cool. I actually use the USB/1394 box I reviewed here, and its fan started crapping out after not a whole lot of months of service. A squirt of oil silenced it, and it's been fine for quite a while now.

Once you've ignored one noisy fan bearing and lost hardware and/or system reliability as a result, you become pretty good at picking the noise, even in a room full of computers :-).

Re:Many fans are fixable

[Anonymous Coward]

Once you've ignored one noisy fan bearing and lost hardware and/or system reliability as a result, you become pretty good at picking the noise, even in a room full of computers :-).

Why don't you just say you don't feel like writing articles over Christmas, instead of making up these bogus excuses. Failed fan. Yeah right :-) (Is it a worry when I know that URL without having to look it up?)

get a used FIA Popnetserver / Iomega p410u

[Anonymous Coward]

I picked one up for 700 bucks. Sure they only come in 400 gig denominations for that, but they are PXEBoot-able... you could remote boot linux and do your own thang.

read your logs...

[Anonymous Coward]

errr... sysadmin 101... read your damn logs...

Re:read your logs...

[Old Uncle Bill]

You guys obviously have a lot more time on your hands than I do. Granted, if it was a unix box, I would probably do that. The term "sysadmin" does not apply to my boxes at home, I just don't care that much. So, you seriously go into event viewer once a week?

For the do it yourself type with a limited budget

[Infinite93]

http://www.mini-itx.com/projects/tera-itx/ A Terabyte capable server for under $300 plus the cost of drives.

Just need to get an old drive enclosure some where (ebay?)

It could even run the drive health scripts itself....

Raidcore is growable

[beernutz]

The Raidcore series controllers absolutely ROCK, and they can to online live growing, raid level migration (say i have a mirror set and add a drive, i can change it to a raid lvl 5, all live and without losing any data) Broadcom recently bought them, so they have great support. It will also let you add multiple cards so when you run out of ports (mine has 8 ports) you can just add another card, and span accross cards transparently.

Re:Raidcore is growable

[TheGratefulNet]

is there solid stable BSD support?

one of the things that impressed me about 3ware was that there are kernel drivers that are production quality. seriously.

I am always interested in good hardware, but it MUST be freebsd-solid. else, its usually a good comment on how poor the hardware is, or how closed the drivers are.

finally, since I've never come across that brand before (and I'm not so junior in pc's, raid, etc) - then I do worry if I got one of these controllers - how long they'll be around and how easy it will be to get replacements.

3ware, promise, highpoint - have all been around for a while and their products (well, except for 3ware) are very commonly available at many stores.

Re:Raidcore is growable

[TheGratefulNet]

from their faq:

Q: Do you have plans to support Novell NetWare, FreeBSD, MAC/OSX, or other OS's?

A: The BC4000 Series SATA RAID Controllers currently support various Windows versions and several Linux versions. Support for other operating systems is planned but not yet scheduled. Any information you wish to provide regarding other operating system needs, the criticality and the timeframe, would be welcome.

BEEP! wrong answer. no bsd drivers.

3ware has 'em.

I have no idea why there's no bsd drivers, but this takes them out of any serious running for any serious server work...

/sigh...

Typical Slashdot

[bhima]

Let me summarize all of these comments: No we Don't Know, Not Really.

Which really disappoints me as I will soon run out of room in my PowerMac with my extra drive bracket.

Is it just me or is Slashdot becoming exponentially more useless?

Re:Typical Slashdot

[bhima]

I have just found this...http://www.engadget.com/entry/1234000570024 663/

Re:Typical Slashdot

[Don'tTreadOnMe]

bhima,

Check out macgurus.com. They have some good solutions that seem pretty inexpensive.

Re:Typical Slashdot

[bhima]

I have seen this...

Plus: It's in my price range, It's in my size range.

Minus : That whole external wiring thing (I have one of those water cooled PowerMacs and they don't like having holes in them), I don't think they ship to the EU, also it seems like a hack

But hey what can I expect for the money! So I'll probably wind up with a variation of this

Re:Typical Slashdot

[Don'tTreadOnMe]

I have a few of these on Windows desktops, which I modified as described here.. It is a hack, but then, if you don't want external firewire, fibre, USB, or SCSI, there's not really any other choice. I haven't found anything cheaper than this, and while it certainly would be an aesthetically non-pleasing solution, it is a solution that seems to fit what the OP requested.

Meanwhile, one click at the macgurus website gave me this:

"International orders are shipped Fedex International Priority or UPS Worldwide Express..."

So maybe you can get it in the EU. :->

Why RAID 1+0 is safer than RAID 0+1

[adb]

Here's a clear and concise explanation, with pictures.

With a striped pair of mirrors, a total failure happens only if both drives in one of the mirrors fail; there are two ways this can happen.

With a mirrored pair of stripes, a total failure happens whenever any two drives in different stripes fail; there are four ways this can happen.

In both cases, there are (4 2) = 6 pairs of drives that can fail. Given that two drives have failed, there's a 2/6 = 33% chance that the RAID 1+0 will fail, but a 4/6 = 67% chance that the RAID 0+1 will fail.

Re:Why RAID 1+0 is safer than RAID 0+1

[TheLink]

Thanks! Got it :).

hang on folks...

[Anonymous Coward]

Are we sure this is SATA and not SANTA? He's watching, you know...shame to miss out on a new Linux box Saturday morning

only one place you need to go

[datazone]

rackmountpro! they are best place to get server stuff, and they treat their customers good. I have purchased a decent amount of stuff from them over the years, and its a joy.
here is the exact link for what you want:

http://www.rackmountpro.com/productpage.php?prod id =2135

Hall Effect Fans

[Detritus]

There are fans that have embedded Hall effect sensors that generate a pulse once per revolution. I've used them on embedded systems. Wire the sensor leads to a parallel port and write some software for a task that counts the pulses from the sensor and calculates the fan's RPM. It can generate an alarm if the fan slows down or stops.

My Setup

[bulldog2260]

I am running a Promise Fast Trak 100 Tx2 Pro, it included the RAID 0/1/JBOD Card, and 2 SuperSwap 1000 enclosures. The card is PATA, and the superswap enclosures will allow you to hot swap the drives. Link to Card

I also have the RAID on the motherboard, but it does not support hot swap.

Running FreeBSD, I have Samba, NFS, and Appletalk, BIND9, Apache (for some testing), Postfix (for relaying the mail), and other goodies.

obObvious comment

[RMH101]

if you build a storage box (NAS, USB, Firewire, whatever), why not put it in a larger box, like a small ATX case? if the fans fail, it's not going to cook anything. The only reason the OP had the failure is he was using a drive in a small case that didn't allow heat dissapation...

Some ready to go hardware choices

[Don'tTreadOnMe]

We have recently purchased some hardware like this to expand both network attached storage, and desktop solutions. We have a GIS department that regularly fills up 400GB drives and were not well backed up, so we needed to get them a terabyte or so of raw storage (500GB mirrored) for their desktops, and multiple terabytes on the network for archiving.

We ended up with a server like these rackmounts, with 24 hot swap drive bays, Windows 200 server license, 4 hot swap power supplies (3 live, one redundant), two 3Ware SATA 12 port cards (so no redundancy in controllers), no drives, for about $6,200. We purchased from biz.tigerdirect.com, who were very competitive, and beat any other price I could find to throw at them, and then threw in a 3 year 24x7 warranty. We got 12 250GB drives from newegg.com, because their drive prices just can't be beat. We went with RAID 5 with one drive as a hot spare, so we have 2.5 TB of storage.

Now, this isn't the fastest horse on the block by any means, but we aren't serving databases or working directly with this data, we just needed gobs of reliable storage. I'm happy with this so far.

For desktop solutions, we ended up with a modification to MacGuru's roll your own SATA RAID. We added the internal enclosure from Addonics so that we could have hot swap drive bays. This also means extra bays, since the enclosure reduces the space the drives take up. Also, it's easier to install these bays than it is to screw rails on every drive. O.k., so screwing rails isn't a big deal, but at work, my time is money, so there it is. We used the Addonics RAID cards, which seem to perform nicely, and these 8 port SATA port adapter (scroll down) let's you connect the external array to your main desktop with minimal fuss.

We're very happy with these desktop RAIDS that operate at a very respectable speed - Our GIS folks need desktop access to terabytes of data for their processing, and we're too cheap to buy them workstations. This has been a very cost-effective alternative for us. Hope this helps.

Here you go...

[Ironsides]

http://www.accusys.com.tw/Acuta/Acuta_web.html

USB 2.0 and FireWire or External SATA connection for 4 drives in raid 0,1,1+0 or 5 set ups.