PDA

View Full Version : Raid Fomatting



SCS
06-23-2004, 12:51 PM
What is the best choice for 'Cluster Size' when formatting a software raid under WinXP and why?

djlithium
06-24-2004, 04:47 PM
The windows XP Default is the best.
why? because its easier for windows to write default values than other values for the cluster size because it has to do more translation work before recording to disc.

mgrusin
06-24-2004, 10:05 PM
It's been said here in the past that the default (4K?) is the best (only?) because the defragmenter included with Windows cannot deal with any other cluster size. That said, I've tried using the defragmenter several times on my array and it never seems to be able to totally defragment it - I think it has problems with large files.

Knowing how disks work but not knowing how Windows works, I would have thought that the largest cluster size available (64K?) would be preferable for the sheer size of the files we deal with. But I have not tried this myself (I will some day), and will defer to the more experienced.

-MG.

wvp
06-26-2004, 01:48 PM
What DJ said I'd agree with. Why? read on...

"cluster Size" is a way of segmenting your hard drive into 'data blocks'. When clusters are larger, fewer of them are needed than when they are small. A 10,000 byte (10kB) file would require three 4 kB clusters but only one 16 kB cluster. However, if you store a 2kB file into a 16kB cluster you are wasting 14kB of disk space. Logic would say "I'll go with the smallest cluster size then" Wrong!
The more clusters per disk, the more you are defeating the performance you get by having clusters used in the first place.
In the case of a video array one can argue to make cluster sizes larger (up to 64kB), as many of your files are huge. This will reduce the clusters and thus keep more of the file from becoming fragmented. However, if you store a project file (which is very small) into one of these clusters you will be wasting HUGE amounts of space.
Changing the cluster size from the default (4kb) is not going to give any big performance boosts, but will give you more disk space.

mgrusin
06-29-2004, 06:57 PM
It's true that larger clusters waste more space for smaller files. But in our sandbox, it may not be a big deal. If you use 64K clusters, and store 1000 1K files, that will take 64MB (instead of 4MB for 4K clusters!) But that's still nothing compared to an 80GB video file.

The real question is: will the system access larger clusters faster than smaller clusters? The reason we're all here is to get video on and off our drives AFAP. RTV uses something like 22MB/S. That's 5,400 4K clusters to write or retrieve every second. If you use 64K clusters, the machine will only need to retrieve 350 per second. That's less work for the system, and if the pipeline isn't already saturated, theoretically faster.

I admittedly don't know exactly how data transfer works in modern systems (my education was ISA bus and ATA drives). But my untested, educated guess is that larger clusters are by some amount faster than smaller clusters. It would be an interesting experiment to try different sizes and see what happens (which I would do if my array wasn't always busy making me money).

BOTTOM LINE: if you stick with the default cluster size, you'll be on the safe, paved road where lots of people have gone before (and more importantly can help you when you get stuck). But it may be interesting to experiment with larger cluster sizes, especially if like me you're trying to squeeze performance out of a slower system.

-MG.

ScorpioProd
06-29-2004, 09:07 PM
I did extensive testing of this long ago and found no tested speed difference with different cluster size on my four Cheetah array.

I use the default cluster size now.