Hi. I sent this email to the list last week, but for some reason I never
saw it show up on the list. I apologize if it appears twice now.
We recently bought two Dell PowerEdge 2900 servers each with a 2.3 Ghz 5140
Xeon, 4 Gigs of RAM, 8 15k SAS drives, and a PERC 5/i raid controller with
256 megs of battery backed cache. Our database is more of an OLTP type, and
everything I've read says that 10 would be better, but I thought I would
test both 10 and 5. The other guys here wanted to run the Raid 5 with a hot
spare, so the RAID 10 uses 8 disks and RAID 5 uses 7 plus the hot spare. We
are running CentOS 4.4. I started testing with bonnie++ 1.03a (bonnie++ -s
16g -x 3 -q ) and I got these numbers. These tests were done with a RAID
stripe size of 64KB, no RAID controller read ahead, and write back caching
enabled.
RAID 10
name file_size putc putc_cpu put_block put_block_cpu rewrite rewrite_cpu
Raid10 16G 50546 94 149590 34 81799 17
Raid10 16G 50722 95 139080 31 82987 17
Raid10 16G 50526 94 148433 33 82278 17
getc getc_cpu get_block get_block_cpu seeks seeks_cpu num_files
50678 84 236858 30 602.7 1 16
50845 85 240024 31 594.8 1 16
50921 85 240238 31 547.5 1 16
Raid 5
name file_size putc putc_cpu put_block put_block_cpu rewrite rewrite_cpu
Raid5 16G 51253 95 176743 40 87493 19
Raid5 16G 51349 96 182828 41 89990 19
Raid5 16G 51627 96 183772 42 91088 20
getc getc_cpu get_block get_block_cpu seeks seeks_cpu num_files
50750 83 232967 29 378.5 0 16
51387 84 237049 31 385.0 0 16
51241 84 236493 30 391.8 0 16
I was somewhat surprised that the RAID 5 was equal or better on almost
everything. I assume this must be because it has 6 data disks as opposed to
4 data disks. The one number that I find strange is the seeks/sec. Is
there any reason why a RAID 5 would not be able to seek as quickly as a RAID
10? Or are the numbers from the Raid 10 bogus? I've also done some testing
with Postgres 8.2.1 and real world queries, and the two machines are
basically performing the same, but those seek numbers kinda bug me.
Thanks,
Dave Dutcher