HDD's

I have never seen anything talking about writing only to one platter inside a drive, before moving to the next platter.

The description I gave, goates, is how that block of data is actually written to the hardware inside the drive. It is the first stage of "writing it across" the drives. It is how the individual stripe is created.

 

Assuming an empty drive, d1, starting with the first writable track, say track 2, (d1t2) at the outside edge of the top surface of the top platter is where the "stripe" is going to begin. The write heads on all of the drives in the, say 5-drive - array are all moved to their track 2 position. If one looks at the arrangement of all of those tracks on each platter inside the drive - one can see that they form a "cylinder" of tracks.

The first segments of the incoming "block" of data - the data stream assembled from cache by the raid controller - will be written to the first available (4k sized block - generally changeable) space on d1t2, and continuing to write until the starting point comes around. The controller then switches to writing data to the underside of the first platter, on its track 2, repeating the write as the disc rotates, all the way around to its start point, then the controller switches the data stream to the top track 2 of the second platter, and so on, using each track 2 of each platter top surface and bottom surface, in the drive, all the way to the bottom track 2 of the bottom platter.

The controller then moves the data stream - the block - to point to the next drive (d2) in the array stack, where it continues the write process on its track 2 (d2t2), again repeating the same write process - and then into the next drive (d3t2). This is what is meant when talking about striping "across" the drives. This is how striping is physically done.

 

Only when the data block size exceeds the sum total space on all "track2's does the read-write arm have to move its position to the next (available) track. When that first block of data is done - the process repeats itself over again, using the next available empty track/cylinder.

The graphics in all of the diagrams we've shown are all simplified overviews of the process, not the detail, with the exception of the strange first part of the NCIX Hitachi demo. Raid 5 is similar, except there will be additional parity data written so that, in the event of a drive failure, a new drive can be swapped in, and the array can rebuild itself. Raid 0 doesn't have this feature.

 

Man! I sure hope this one is done to death by now! :lol:

 

You can understand why I, too, want to replace my hdd's with SSD's. So much simpler!

Man! I sure hope this one is done to death by now!

 

What you're describing is not the overall concept of RAID 0. That may be how platter hard drives handle data internally, but it is not describing how RAID 0 works. Perhaps two different uses of data striping are being conflated here. Under RAID 0, the data is split evenly between all drives in the array and written to each drive simultaneously. Reading files back is just the reverse with read operations spread across all drives in the array. This is where the major speed increase comes from, and is what all of those links I've posted above are talking about. It does work with SSDs, and very likely with other drive technologies in the future. No need for platters at all.

 

In fact, even Intel thinks RAID 0 can be done with SSDs.

 

http://www.anandtech.com/show/6161/intel-brings-trim-to-raid0-ssd-arrays-on-7series-motherboards-we-test-it

 

So, at this point I'm just going to agree to disagree.

I can't resist:

 

PaulJ's initial RAID0 description is too detailed and the forest got lost in the trees, though it is correct considering that the described operation is performed concurrently on all drives in the stripe set required to complete the write. But also any decent HD disk controller will attempt to minimize seeking while performing mulitple sector writes, regardless of whether the disk is in a RAID0 stripe set or not. As for RAID0 being inappropriate for SSD, it's not if massively high bandwidth realtime data acquisition is your goal; but then again we're now talking about serious computing, not a hobbyist pasttime.

I can't resist:

 

The point I was debating was Paul's assertion that RAID 0 does not apply to SSDs.

 

Sorry, Wolf, don't even think of RAID 0 as any kind of solution when using SSD's.......

 

...

 

RAID 0 does not apply to an SSD. RAID 1 would work, but at a high cost - at present prices - with little gain in either reliability or performance.

 

My point is that it does, but doesn't really offer much for FSX or most users. As RAID 0 support is included on even sub $100 motherboards it isn't just limited to high end computing any more, and questions like the OP's come up more often now.

I missed that point on SSD raid appropriateness in your agurments, which BTW I did address in my post. I do agree that PaulJ did missed the mark on the basic concept of RAID0 striping: to utilize concurrency to speed up read/write operations.

Aaaah, but concurrency was never taught (to me many years ago) as a part of a Raid 0 array. Basically a single stack of disks written in a stripe, top to bottom: I suppose it doesn't take too much imagination to make a controller (or two or three) do that with several stacks concurrently. However if I had that much data to spread over that many drives - I would be using Raid 5. (sigh) One is often behind the eight-ball. :(

 

Thing is, we (I) never used anything here but 1 for a server and 5 for the systems drives. This is the type of stuff I worked with, plus later and earlier sans. But never once did 0. The old HSG80 arrays are entered manually, disk by disk, EVA3 and 4000 by simply answering a script with a (large) MB figure and raid type.

 

Anyway - it was a good discussion. I learned "a bit" and now you really know how old I am! :lol:

However if I had that much data to spread over that many drives - I would be using Raid 5.

 

Well, RAID 0 really isn't RAID as it lacks any sort of redundancy. It's purely about speed and is mainly used for video editing or other situations where maximum throughput is required. Certainly not something that would be commonly used in a data center where data protection is critical.

 

I apologize if I came across a bit too argumentative. It certainly wasn't my intention.

 

For fun, try doing a search on RAID 0 arrays and USB memory sticks.

 

http://analogbit.com/node/4

 

http://www.techpowerup.com/forums/showthread.php?t=114613