Ok so the common wisdom here says that with SSD's you should use noop or deadline for your scheduler and not CFQ. However I saw that as far back as 2008 CFQ can recognize a SSD and act accordingly...


CFQ has some optimizations for SSDs and if it detects a non-rotational
media which can support higher queue depth (multiple requests at in
flight at a time), then it cuts down on idling of individual queues and
all the queues move to sync-noidle tree and only tree idle remains. This
tree idling provides isolation with buffered write queues on async tree.

This is taken right out of the kernel txt files.

My question is has anyone actually bench marked and tested which one is in fact best/fastest for SSD's?

Thanks

Not real benchmark here, but using samsung EVO SSD and hdparm -t test, I found Deadline scheduler slightly faster than other two, but it was by a very few margin

Here's a benchmark from a while back:
http://ubuntuforums.org/showthread.php?t=1464706

In my tests I got the same results regardless of media being SSD or HDD, deadline is the best. In fact, it is also best regardless of filesystem although I haven't tested all filesystems.

Hi,

If you match the 'SSD' to your system via optimization so that the cache & write back are functional(if the SSD supports write back) then noop works fine for me. If you are in need of database management then use 'deadline'.

Get your 'SSD' specifications then optimize your GNU/Linux based on that. Most modern kernels can be optimized for 'SSD' operations. Setup the specific device for a specific scheduler, do not make the scheduler system wide. Assign the scheduler to that device so you will not create a problem.

Be sure to do a LQ Search since this subject has been covered many times.
Hope this helps.
Have fun & enjoy!