I'm using couple of arrays made by linux software raid1 and raid5. raid arrays are non-uniform and they cover about 60% of disk space. everything else is not protected by redundancy.
Personalities : [linear] [raid1] [raid6] [raid5] [raid4]
md4 : active raid5 sdf1 sdo2 sdn2 sdm2 sdl2 sdk2 sdj2 sdi1 sdh1 sdg1
443658240 blocks super 1.1 level 5, 64k chunk, algorithm 2 [10/10] [UUUUUUUUUU]
bitmap: 0/1 pages [0KB], 65536KB chunk
md2 : active raid5 sda3 sdi3 sdh3 sdg3 sdf3 sde2 sdd2 sdc2 sdb3
11301533184 blocks super 1.1 level 5, 64k chunk, algorithm 2 [9/9] [UUUUUUUUU]
bitmap: 0/11 pages [0KB], 65536KB chunk
md3 : active raid5 sdj1 sdp1 sdo1 sdn1 sdm1 sdl1 sdk1
1988476416 blocks super 1.1 level 5, 64k chunk, algorithm 2 [7/7] [UUUUUUU]
bitmap: 0/3 pages [0KB], 65536KB chunk
md1 : active raid1 sdb2 sda2
52339648 blocks [2/2] [UU]
bitmap: 5/200 pages [20KB], 128KB chunk
md0 : active raid1 sdb1 sda1
96256 blocks [2/2] [UU]
bitmap: 0/12 pages [0KB], 4KB chunk
unused devices: <none>
I'm doing complete check on all arrays every month
echo check >> /sys/block/md0/md/sync_action
echo check >> /sys/block/md1/md/sync_action
echo check >> /sys/block/md2/md/sync_action
echo check >> /sys/block/md3/md/sync_action
echo check >> /sys/block/md4/md/sync_action
If any unreadable sector is found, raid will try to resync it resulting with sector reallocation. so far so good.
fans used inside the case are not ideal - they are S12 and they lack pressure needed at least on CPU heatsink. I already had them. but, considering low CPU load and heat dissipation, they are doing good job. core temps top at ~58C when both cores are loaded with prime95, and core temps cannot reach 50C during normal operation (pbzip, direct connect hashing, compiling...).
fans on cages are good - they are P12.
fans in front of lower drives are NF-B9, three of them.
about Samsung F2...
well, it's not blazing fast. nor it's fast at all.
but, it's quiet, cool and cheap, so it feets my needs: massive storage without too much need for random access and some fancy seeks.
sequential read is good - it reaches 1.2TB/s during raid check. it's not too fast considering every of 16 drives gives 75MB/s, but it's not bad eighter.
to achive that I had to put additional controllers in PCI-E slots connected to north bridge.
my monthly average disk read including complete check of all arrays is 9.5MB/s (sum for all disks).
average write is ~700KB/s.
average temperatures for Samsung are 28 to 30C.
btw, Samsung F2 is known to have terrible multitasking algorytm. I've tried to lower impact of that by making sure two most busy arrays don't share single phisical disk (md2 and md3) and by increasing use of write buffers (commit=29, commit=30, commit=31... as mount options and vm.dirty_* tunning).