When using a raid 4/5/6 with mdraid there is the stripe_cache_size option,
to speed things up, especially writing.
When using LVM, you can use the mdraid backend as well, but you can not set this using the lvm frontend.
After digging around a lot in the lvm git repos, I found out that the developers seems to planned those options but they are never implemented. After more digging, I found out that you can set those options using the device-mapper.
I'm not totally sure if this is the best way but it works. unfortunately not persistent.
First you get the current settings from dmsetup:
$ dmsetup table
[...]
kujira-srv: 0 11525947392 raid raid5_ls 3 128 region_size 8192 3 254:5 254:6 254:7 254:8 254:9 254:10
[...]
So now you unmount the lvm and reload with new arguments, as well as resume the device:
$ dmsetup reload kujira-srv --table '0 11525947392 raid raid5_ls 5 128 stripe_cache 16384 region_size 8192 3 254:5 254:6 254:7 254:8 254:9 254:10'
$ dmsetup resume kujira-srv
$ dmsetup table
[...]
kujira-srv: 0 11525947392 raid raid5_ls 5 128 region_size 8192 stripe_cache 16534 3 254:5 254:6 254:7 254:8 254:9 254:10
[...]
In the kernel.log we can see now:
[ 1463.958469] md/raid:mdX: device dm-6 operational as raid disk 0
[ 1463.958472] md/raid:mdX: device dm-8 operational as raid disk 1
[ 1463.958473] md/raid:mdX: device dm-10 operational as raid disk 2
[ 1463.958867] md/raid:mdX: allocated 3316kB
[ 1463.959037] md/raid:mdX: raid level 5 active with 3 out of 3 devices, algorithm 2
[ 1463.959039] RAID conf printout:
[ 1463.959039] --- level:5 rd:3 wd:3
[ 1463.959041] disk 0, o:1, dev:dm-6
[ 1463.959042] disk 1, o:1, dev:dm-8
[ 1463.959043] disk 2, o:1, dev:dm-10
[ 1463.959293] created bitmap (344 pages) for device mdX
[ 1464.001044] device-mapper: raid: 16384 stripe cache entries
[ 1472.939823] mdX: bitmap initialized from disk: read 22 pages, set 0 of 703488 bits
The write speed is now actually faster.
By the way: The current jessie kernel 3.16 will crash if you try this! At least in stretch it is fixed (4.8).