Fucking workhorse.. This old dual Xeon box

That's interesting. Thanks for posting, MrMaxwell...

Someone here is capable of interest...
That's interesting..
Thanks for making a point, a point to be pondered

Welcome to the twilight zone.

Still using my dual Xeon with 8 ultra scsi 360 15,000rpm hard drives. Rock solid and gets the job done.

Xeon's are some badass processors. Intel got one right with them.

Raid 0 for paging is fine.

Raid 0 for anything else is retarded. You shave off a couple of MS to risk X% of your filesystem missing, where X is usually around 50%, since most folks just RAID0. I might consider JABOD, but I'd still run (my disks) as RAID5 or separate. RAID0 is not your buddy. RAID0 is not your pal.

I'm very new to raid but I like it

Just a bunch of disks, it becomes a big mess fast... doesn't it?
And raid 5 is two raid0s mirrored with a spare in case of a failure?

I know that raid 0 is volatile but you probably should just treat it like any other drive.. drives fail.. so to me two as one is just one drive .. maybe think of that drive as being made by some off brand because it's half as trustworthy
Is it really 2ms difference in the seek time? That's significant!

Not quite.

RAID0 is just 'treating it all as one disk', basically. RAID5 segments different drives (at least two) and uses another for parity. If one fails (and eventually, they all do), you can restore all of the data by the pre-existing content that is there and the parity. In the event the parity drive fails, it just rebuilds that. So, it'll take you 3 disks to make 1, but when one dies, you're only slowed down - you're not fucked.

You can't use 'two as one' as an analogy because EVERYTHING now has at least twice the amount of failure probability - two spindles, 2 controllers, (and it should be 2 different power rails, at least), etc, etc.

RAID0, if you have 2 drives, one fails - there went half the filesystem. JABOD also suffers from this, but JABOD isn't quite as bad; RAID 0 requires everything be matched up size-wize - JABOD you can just throw anything together - and unless you totally fuck up, you may be able to stitch it back together. Some filesystems can 'heal' and work better with a JABOD system - ZFS and Reiser are semi-capable of this.

At the very least, I'd suggest RAID1, where it's just a mirror of the same content spanned across two drives. One dies, hope to hell the other one isn't.

On average, RAID0 does lookup faster, and that average with 360u SCSI w/ huge cache was about 2ms. With SAS/SATA, it's probably similar. May shave 4 seconds off of the wait time for shit to load, and you just lost your life's work. What a tradeoff!

I'm no math guy but I'm still not seeing how my analogy is so flawed.. if it's twice as likely to fail, it's half as reliable, right? If RADI0 is treating everything as one disk, could you use four as one to make it even faster? How many scsi drives would you generally need to be using to hit the wall and max out the system bus?

I think basically what you're trying to say is not to trust RAID0 for storage, right.. ? I gathered that, I'm just looking at the speed and performance aspect of it, myself..
What would it be called if you had two raid 0 arrays mirrored? That'd be a more reliable setup, right, but not necessarily "twice" as reliable, though, right?

http://www.pcper.com/article.php?aid...=expert&pid=10
These ram drives are amazing but it seems like they're even less reliable than raid 0
I don't know much about SSDs yet, need to read up on them.. but these ram drives are amazing if they provided enough storage, they'd be great to have your OS/paging file on.. If I got one I'd figure out a way to have it back itself up to something every few hours .. they have a battery backup but that still isn't too comfortable

I know nothing about computers, really...but mine rocks

I have a Dell Precision Workstation 670 with Dual Xeon Processor (3.0's) with 4GIG of RAM. It was built in 2004. At that stage in the history of PC's this thing had the ability to change the rotation of the Earth.

It's a workhorse...I love it

buy 1-2,000,000 uniques a week and see how it does

not sure but I think ppl are referring to having a personal dual xeon box. for rendering, in-house file server, etc.

I also have a dual xeon server in my house for such purposes, it is an awesome workhorse, performs better than any computer I have ever owned, and I forsee many more years of flawless performance.

love it

That sounds more like RAID3, which has a single dedicated parity disk. RAID5 stripes parity (as well as data) across all of the disks in the array.

This is RAID0+1, which is striping two drives, then mirroring the stripe. It's better to do it the other way around (RAID10): mirror two drives, then stripe the mirror. Speed and performance wise there's little difference between the two, but a 4 drive RAID10 will rebuild twice as fast than a 4 drive RAID0+1 if a drive has to be replaced.

RAID0+1/RAID10 offers the speed benefit of striping, and also better average seek time if your controller is smart enough to properly load balance by sending a request to the drive whose head is likely to be closest to the sector it wants to access. Obviously RAID0+1/RAID10 also offers the benefit of mirroring so a single drive failure will not kill your array. It can actually handle one more failure depending on the position of the drive in the mirror. (After the first drive fails you have a 50% chance of losing the array when the SECOND drive fails)

Anyway... RAID is for uptime, not for reliable storage. You need a backup for that. :D If you're comfortable with it, you can use RAID0 or JBOD for your backup volumes, since they don't technically need to be fault tolerant. Depends how paranoid you are, and how important your data is.


BTW... Pentium Ds... old Xeons... with the amount of power those things take you might actually save money by upgrading to something more efficient. I replaced a Pentium D 830 with a Celeron E1400 that consumes about half the power. If that box ran at 100% load 24/7/365 then it would save something like $USD75 worth of electricity in a year. Of course it doesn't run at 100% load, but it's still saving at least

With regards to reliability an all RAM drive used for paging should in theory be more reliable (and significantly faster) than a mechanical hard drive. If you use it for paging only then retention of contents when you power down isn't an issue, in fact some may consider that a security benefit.

If your OS supports "excessive" levels of swap then using an addon RAM drive for swap effectively increases your available system RAM... it can address (in a roundabout way) more than your mainboard can physically support. It won't be swapping at native RAM speeds because of the overheads of the PCIe bus, SATA interface, etc, but it's still a hell of a lot faster than a mechanical disk. (I don't think this trick would work with a 32 bit OS because of the 4GB limit)

I seem to recall seeing a 5 1/4" bay SATA version of the iRAM, or perhaps it's another name made by another manufacturer. I was looking into this a little while ago, but for the price it may be worth considering just upgrading RAM and/or mainboard...

i have been using raid 0 for my main system for years right now running 2 36 gig 10k rpm raptors my next system will be 2 15k rpm sas drives in raid 0

Serial attached scsi, right?
They put them in those nice high end new servers, right?

I don't know alot about SSD drives.. are they almost as fast as iram?
I know that paging from an iram wouldn't be a problem, what I meant was, I'd like to have something big enough to hold the O/S and all of the programs, too.. Perhaps I could page from iram and put the os/programs on an SSD drive? That seems to make sense.. The SSDs are alot faster than mechanical scsi, right?

When the controller knows how to balance requests, is that similar to NCQ on sata drives?
I think NCQ is in the drive itself and it prioritizes the buffer for more effecient reading and writing... would be great to just have it in the controller, guess you'd have to for raid0

NCQ is the drive deciding the best way to order requests coming from the controller, generally to minimise head travel over the platters. It improves performance under load (ie more than one request outstanding), but in a mirrored setup it's better if the controller decides which drive to send the request to in the first place... drive #1 and drive #2 can't communicate and figure out which drive probably has its head closest to the requested sector, that's up to the controller...

SSD are improving but they're still very expensive. Some of them have trouble with writing because flash can only delete a relatively large block (such as 128k), so even writing out a single sector of 512 bytes will require reading in the 128k, deleting the block, modifying the 512 bytes, then writing out the 128k again. Newer hybrid versions get around this with an onboard RAM cache that bunches up write requests to get them onto the flash more efficiently.

I've never heard of a pure RAM drive in significant capacities, although I'm sure it's been done... the cost would be huge, even compared to SSD... for example 64GB worth of DDR2 Kingston RAM sticks would cost $USD1100+...

Do they make ram drives with 64gb capacity, or more than four slots, now?
How big can you buy a ram stick these days? 8gb or some shit? I'm so behind the times over here.

I would definitely agree that I'd much rather have an actual controller making the decisions. It knows before the drive knows. The controller is just the same thing as NCQ but it's done before it's even sent to the drive, yeah?

If SSDs can only be formatted to 128k blocks... that fucking SUCKS
NTFS you only go up to 8k blocks, right?

I'd also like to know more about the best size blocks for a striped array..