Stripe Size & positioning performance

simple question here, it is said that smaller stripes decreases positioning performance.... well what about if the stripe were so small that the positioning was identical???

just to clarify, you are talking about RAID 0 correct?

I'm not sure I get your question. Decreasing stripe size increases transfer performance (because the files are spread across more disks), but decreases positioning performance (takes longer to find the smaller pieces). (assuming, as adampi said, that you are referring to raid0)
http://www.pcguide.com/ref/hdd/perf...rfStripe-c.html discusses this a little. If this isn't helping, try to clarify what you're asking.

sorry i took so long to get back.. yes im talking about raid 0, and ive already read that page! haha...

so what happens if u make the strip size really really tiny? (say one byte if that is possible) theoretically, could not all the disks be in the same position - therefor eliminating the seek time degridation?

Still not sure what you mean by "all disks in the same place"
You can't write more than one thing to a byte. Decreasing the stripe size will decrease positioning performance. Its just that simple. The smaller your stripes, the more likely a file is to be spread across more than one disk, causing seek times to increase, because two or more disks each have to find portions of the file before it can be read/written to. If you had some theoretical tiny stripes, then a two letter file containing just "ab" would be split, with "a" on one disk, "b" on the other. This would obviously slow things down.
I'm still not sure if this addresses your question, but HTH

(^^;?(Well, Well, you seems to repeat same paradigm. You'd better begin with explain what you understand now.)

When we have a disk formatted as allocation Unit size 8kb that is cluster size 8kb in old term. It means that any small files on Windows have to have 8KB at least and requires or consume memory for it.

Then if we set the stripe size to 4KB, 2KB, 1KB. It means that any files on Windows is splitted or sprited. When we set it to 8KB or more, some files would not be splitted.

Note: Most of us are rather practical than theoretical. It is reported that IBM drive makes better performance at 64KB or 128KB and Quantum does at 4KB.

nope, u still dont understand the question i dont think..


i take the seek time degredation to mean this: bunches of data are stored in a series on one disk, often this means one disk can return all the requested data, which means while its head is moving around - the other disk remains stationary - and so once the next read begins, the position of the heads on both disks are now offset by however much data the last head read - causing a wait for the other head to catch up?; now what if you have a 1 byte stripe so that both heads are always in the same relative position? - shouldnt the result be: no change in seek time, but still an increased data transfer rate?
(yes i mean byte not kb)

i hope that is clear enough...

Byte level striping is possible and done (RAID 3 for example), but it has its downsides as well.
Read this and see if the descriptions and discussion help with your question.

(^~; ( As Karsh seems not to bear company with you seriously. )

If you understood what I said, I suppose you have already noticed.

When the cluster size is 8KB and you have 4 drives, then you set the stripe size 2KB, at this point all files are supposed to be striped. Here comes Turning Point, the more you set the size small, the more you would cause delay, because after this point you are supposed to increase seek-read-write cessions in vain until its limit 1byte. In fact, it has nothing to do with raid directly, copying one 100kb file is much faster than copying hundreds of 1kb files. You confounded this.

Then, Here's simple question. Which is faster, copying 100kb file on standard HDD or copying one hundred 1kb files using stripe size 1KB on 2 drives?

despite my drunkenness i wonder if perhaps if i shall still make some sense here, but if i am to take this comment directly it seems that you would say raid0 is completely worthless? smaller = worse and bigger = better then why not just have it as big as possible which would be the entire capacity of the drive?.......

and karsh, i didnt see anything even remotely related to my question on your link

(^^;?(Everything has its merit and demerit. The important thing is to decern its characteristics, I think so.)

No it is not worthless. You should get hold of the fundamental point. Raid0 can handle the multiple drives at the same time against one file requested from OS. It is striper's pride. When you set the size 1byte and prepare drives 8000 pieces, Windows regard them single drive of extreme high performance.

It is the matter of combination in the end. When you set the size to entire drive, then no file is splited and no operation is possible. Quite natural, isn't it? Although it is unclear how windows regard the drive conditon.

It then should be noted that RAID is a contrivance or an idea to try to make use or reuse of cheap drives as one high performance drive from the start. Then many people misunderstood it as if it were somewhat a new miraculous high tecnology. Of course, it was an excellent idea, but it still has some defects and problems. Pay attention to the word "CONTRIVANCE". I'm thinking the word is the key to all device structures as well as raid.

I posted that mostly because of its mention of RAID level 3, which uses byte level striping, as you had been proposing. There was also a section on positioning vs transfer performance again.
I don't know how else to explain, so I'll leave it up to megumi.