Depending on what caches are enabled in your storage, this can have reduce disk write throughput. Even when you have a battery backed write cache, unless the batteries can be hot-swapped, it is likely that at some point in time your server will run in production with the RAID write cache disabled.
I have a test server that uses HW RAID 10 with disk caches disabled. I disabled the battery backed write cache on HW RAID and ran sysbench fileio tests using ext-3 and XFS. The server uses CentOS 5.2. The numbers below repeat results I gathered a few years ago after browsing ext-2 source code to confirm the behavior.
Writes-per-second
threads 1 2 4 8
xfs 162 294 501 800
ext-3 161 162 161 165
Notes
Script to run sysbench. It was run for 1, 2, 4 and 8 threads for 30 seconds.nt=$1 secs=$2 args=$3 ./sysbench --test=fileio --file-num=1 --file-total-size=4GB \ --file-test-mode=rndwr --max-requests=0 \ --file-fsync-freq=0 --file-extra-flags=direct \ --num-threads=${nt} --max-time=${secs} ${args} http://www.facebook.com/notes/mark-callaghan/xfs-ext-and-per-inode-mutexes/10150210901610933
Nessun commento:
Posta un commento
Nota. Solo i membri di questo blog possono postare un commento.