|
|||||||
| Hardware Discussion & Support Discuss your computer - its components or ANY hardware, past/current/future you want, or ask our forum experts if you have a general problem with your hardware. |
![]() |
|
|
Thread Tools |
|
|
#1 |
|
Number Nine
|
RAID 0 questions
1) Does the cluster size selected during format have impact on RAID 0?
2) Do cluster size and stripe size need to match in any way? 3) What impact does stripe size have on RAID 0 performance? |
|
|
|
|
|
#2 | |
|
Obvious Closet Brony Pony
|
I always leave the windows cluster size as default (i recommend you do to)..
however stripe sized/block size... (stripe length?) hugely impacts performance, i'd suggest 128kb if you can, 256kb at most, but no smaller then 64kb.
__________________
Quote:
|
|
|
|
|
|
|
#3 |
|
Member
Join Date: Mar 2003
Posts: 5,989
Rep Power: 69 ![]() ![]() ![]() ![]() |
Stripe Size from the RAID setup is the size of strips that the RAID subsytem handles, in it's caching algorithm. RAID stripe size is applies to the whole RAID array, To change the stripe size you have to wipe the whole array and reset it, losing all data on the RAID array member drives.
Cluster Size for the NTFS file system is what Windows uses to allocate disk space. Windows allows different cluster sizes for different partitions. Different cluster sizes, Different partition types, Different file system formats, all can be used in a RAID array. RAID stripe size and NTFS's cluster size are totally different things. Irrespective of what some may say, there's no real throughput improvement to be gained by trying to match stripe size to cluster size. The stripe size value that you will enter during the RAID setup should be based on the planned drive usage... 64K-128K Good for general purpose strip size, Best performance for most desktops and workstations. Most of the time the hardware is optimized for the default settings or the defaults are what the hardware works best at. Optimum stripe size for most applications would be 64K or 128K... That's why most controllers will default RAID0 to use a 64K or 128K stripe size. 16K Best for sequential transfer (also best for benchmarking, see more info below)... Small stripe sizes work best with benchmarking tools, but that is misleading; these test-tools do not mimic real-world application access patterns; they do random accesses of small segments, and give a good indication of a disk drive + controller's raw performance capabilites, but a poor indication of real-world system performance. More and a better info on the Stripe Size - http://faq.storagereview.com/StripeSize |
|
|
|
|
|
|
|
Number Nine
|
thx for the replies and PangingJr thx as always for the detailed clear response
|
|
|
|
|
|
#5 |
|
Member
Join Date: Mar 2003
Posts: 5,989
Rep Power: 69 ![]() ![]() ![]() ![]() |
no problem, Chaos.
that's my correction of answers to these questions, just try to make it to be more shorter and hopefully more understandable. |
|
|
|
|
|
#6 |
|
DriverHeaven Extreme Member
Join Date: May 2005
Posts: 6,794
Rep Power: 0 ![]() ![]() |
Cluster sizes on NTFS volumes aren't usually left best at default. If you're going to have a partition with LOTS of small files you may want to set the cluster size lower to avoid alot of slack space waste due to allocation an entire 4K cluster to a 100 byte file.
Generally its reccomended that partitions 0-10GB should be 512 or 1K clusters (512 for alot of smaller files and 1K for more general purpose use). Anything from 10 - 50 should be 2-4K again depending on the average file size. Anything 100GB or more should be at least 4K, 8K only if its going to hold files that are exclusively 100MB or more in size 90% of the time. Though it doesn't make too much of a difference in this day and age of super speed hard drives, it can make a difference when trying to cram stuff onto your hard drive and access times will also decrease too (at least in my experience). |
|
|
|
|
|
#7 |
|
Member
Join Date: Mar 2003
Posts: 5,989
Rep Power: 69 ![]() ![]() ![]() ![]() |
Cluster size settings only reveal themselves in actual, literal testing of specific and priority activities. there are too many things that vary to make a prediction of what benefits you in one way or another. For instance, if you make cluster sizes really large, you may find that most of your data doesn't fill the cluster and you effectively throw away a lot of the storage capacity of the drive. If you make the clusters too small, then you increase the amount of read/write activity to the drive map table, plus leave yourself open for tremendous fragmentation on the drive due to all these tiny places that can be scattered.
|
|
|
|
|
|
#8 |
|
DriverHeaven Extreme Member
Join Date: May 2005
Posts: 6,794
Rep Power: 0 ![]() ![]() |
Actually smaller clusters can reduce fragmentation, it only becomes an issue when you've got less than 25% free drivespace.
Besides the MFT bitmap gets fragmented ANYWAY and this doesn't exasperate the problem too much. If you've got a proper defragmenter it makes it all moot, run boot defrags once a month then run your regular defrags often and your drive will be fine |
|
|
|
![]() |
| Bookmarks |
| Thread Tools | |
|
|