I've benchmarked two different resync speed settings in several configurations by test-installing a base system with debootstrap and determining the running time of the package installation phase from dpkg.log. All tests were run on an Athlon II X4 640 with 4 GB RAM.
The first test setup consists of a md raid-1 over two disks. The second and third test setups use a raid-1 that spans two partitions of the same disk in order to simulate a raid of slow disks. Additionally, I've reduced available memory to 256 MB for the third test to simulate installation on a machine with little memory. These are the results:
old: /proc/sys/dev/raid/speed_limit_min is 1000
new: /proc/sys/dev/raid/speed_limit_min is 0
md raid-1, two disks
old: 170s, new: 168s, improvement: 1%
md raid-1, two partitions on same disk
old: 284s, new: 270s, improvement: 5%
md raid-1, two partitions on same disk, only 256 MB RAM available
old: 321s, new: 280s, improvement: 13%
As it may be expected, the gain introduced by setting speed_limit_min to zero is more pronounced with a slow raid and with little memory available for I/O caches.
I believe the patch is low-risk and thus suggest to apply it. I'll do an integration test when it's included in the daily build.
Cheers,
Thiemo