Thread: RAID optimization and terminology: stripe vs. chunk vs. cluster vs. block

hello all,

been banging head on problem last couple of hours , continue turn conflicting information wherever look. raid experts out there can clear things me? i'm trying figure out difference between various raid terminology: stripe size, chunk size, cluster size, block size.

current understanding of raid, stripe made of chunks. chunk size determines various performance characteristics depending on type of data working , average i/o operations. chunk size , number of drives (stripe width) determines stripe size (basically: chunk size * stripe width = stripe size).
cluster collection of blocks, cluster size determined @ file system level , block size determined @ physical hard disk level. make things confusing, cluster size , block size meant mean same thing...
think chunk size raid controller , cluster/block size operating system (assuming hardware raid). seems optimal size these 2 parameters counter intuitive. video, seems want smaller chunk size, larger cluster/block size, correct?

this article tomshardware.com seems conflict this other article zdnet. think zdnet article correct, smaller chunk size = better performance large data such video, thoroughly confused after research. pdf xyratex helped illuminate things, think terminology quite confusing , googling around don't appear 1 confused...

appreciated,

thanks!
zephyr

don't use raid. more ram , won't need screw around raid parameters , raid failure. modern sata disks have enough throughput , enough buffering, increase in disk throughput through raid countered increase risk in failure.

large disks suffer long rebuild times when raid has degraded--sometimes several hours. failure during rebuild time can mean complete raid failure @ point.

unless network 10 gigabit, network latency kill raid benefits.

terminology, hardware raid controllers optimized range of chunk size , stripe layout. in dell poweredge 1750 server perc4/di raid controller, have room 3 disks striping across 3 disks (typical raid5 configuration) , in linux using megaraid driver. there configuration tools tweak raid performance , don't recall specs little googling:

http://www.amazon.com/dell-poweredge.../dp/b00456y41g

host data transfer rate : 532 mb/s
drive data transfer rate : 320 mb/s
maximum size of i/o requests :6.4 mb in 64 kb stripes
maximum queue tags per drive : many drive can accept
stripe sizes : 2 kb, 4 kb, 8 kb, 16 kb, 32 kb, 64 kb, or 128 kb

hardware raid, depends on hardware, obviously, fakeraid , software raid, have perform testing find optimum settings.

in above perc example, disks u320 scsi, can't expect more 320 mb/sec , typical 15,000 rpm drives can put out ~70 ~100 mb/sec due physical limitations. card can transfer 532 mb/sec motherboard. ram tested out ~3,000 mb/sec recall. 3 disks in stripped configuration, looking @ 70*3 or ~270 mb/sec sustained write speed. can tweak stripe size it's broad curve , depends on workload.

in perc example, think chunk size 6.4 mb maximum using 64 kb stripe size. don't know if there optimum.

there's decent collection of performance tools can install:

sudo apt-get install phoronix-test-suite

ran bunch of performance tests, don't have results handy. did find single drive test:

/dev/sdb1:
timing cached reads: 1364 mb in 2.00 seconds = 681.81 mb/sec
timing buffered disk reads: 226 mb in 3.01 seconds = 75.20 mb/sec

so, answer question, there no standard nomenclature either raid configuration or performance parameters. there accepted , commonly used terms. selection of optimum parameters can come extensive testing using disks, raid controller , representative workloads.

after little more searching found thread:

http://www.thegeekstuff.com/2008/07/...h-screenshots/

definitive guide says:

keep default settings!

Forum The Ubuntu Forum Community Ubuntu Official Flavours Support Hardware [all variants] RAID optimization and terminology: stripe vs. chunk vs. cluster vs. block

Ubuntu