I wanted to benchmark my 7410 running RAID 5 with 4 120GB Western Digital WD1200JB (Models: 75EVA0, 32EVA0, 00DUA3) on a Supermicro 370DLE in a dedicated 64-bit slot running at 33MHz. I thought I’d compare those numbers with my new 7500-4LP and see what the differences are. I tested with 2.6.10-ac10, 2.6.11-ck2, and later with 2.6.11-ac6. I recently discovered several people were having high load issues with 2.6.11 and -ck, which might explain the inconsistent numbers I received when testing with -ck. I later decided to completely ditch -ck for server use. The server using an increased read-ahead buffer as suggested by 3Ware. The optimal size I found to be 1536 on this box.
All tests were performed using bonnie++ using -q -f -x 5 -u 0 -m nebula -s 3G:64K -d /home/shared on the system in single user mode with no other activity.
First, the 7410 numbers. As mentioned above, the numbers with -ck2 seem inconsistent. The numbers with -ac are better.
nebula:~# cat 2.6.11-ck2-nfs4.out | bon_csv2txt Version 1.03 ------Sequential Output------ --Sequential Input- --Random- -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP nebula 3G:64k 44943 31 18851 16 47983 19 99.1 3 nebula 3G:64k 30633 20 19846 17 48842 18 97.6 3 nebula 3G:64k 34129 24 20130 17 51255 19 90.2 3 nebula 3G:64k 27469 19 19205 17 48829 19 98.1 3 nebula 3G:64k 33236 22 17465 15 50421 19 82.3 2 nebula:~# cat 2.6.10-ac10.out | bon bon_csv2html bon_csv2txt bonnie bonnie++ nebula:~# cat 2.6.10-ac10.out | bon_csv2txt Version 1.03 ------Sequential Output------ --Sequential Input- --Random- -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP nebula 3G:64k 28435 21 18514 17 60699 25 136.2 4 nebula 3G:64k 25227 21 18478 16 43567 18 151.0 5 nebula 3G:64k 26212 20 18172 16 49778 20 131.5 4 nebula 3G:64k 24929 20 18732 17 53058 22 151.6 5 nebula 3G:64k 24779 20 18752 17 49625 21 170.3 6
Now we have the numbers for the 7500-4LP. Again, we find -ck to be inconsistent with regard to sequential output. The numbers are better with -ac. Although I didn’t test again with 2.6.10-ac10, I imagine the numbers with 2.6.11-ac6 are within the realm of what I might have found using that kernel.
nebula:~# cat 2.6.11-ck2-nfs4-7500.out | bon_csv2txt Version 1.03 ------Sequential Output------ --Sequential Input- --Random- -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP nebula 3G:64k 55041 38 17120 15 49181 19 108.6 3 nebula 3G:64k 38089 26 16991 14 49805 19 108.1 3 nebula 3G:64k 37913 26 16511 14 51549 19 87.9 3 nebula 3G:64k 31405 22 16285 14 49670 19 108.3 3 nebula 3G:64k 34692 24 16726 14 53287 20 84.5 2 nebula:~# cat 2.6.11ac6 | bon_csv2txt Version 1.03 ------Sequential Output------ --Sequential Input- --Random- -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP nebula 3G:64k 39794 27 20615 18 90465 36 112.5 4 nebula 3G:64k 35960 24 18917 17 75453 30 115.0 4 nebula 3G:64k 35500 23 18616 16 55861 21 113.8 4 nebula 3G:64k 38256 25 17831 16 50645 20 110.6 4 nebula 3G:64k 37125 25 18119 16 69135 27 110.8 3
We arrive at an average sequential write speed of 37328K/sec under 2.6.11-ac6 for the 7500-4LP versus 25917K/sec with the 7410. It seems 3Ware’s R5 Fusion technology has indeed increased my performance. The read speeds are far more variable, but the 7500-4LP’s worst sequential read is still better than the worst 7410 sequential read.