The short of it is that the settings on different OS/Distro/AMI
settings for the read-ahead are fairly large (megabytes) and in a
system with random access and small documents (in KB) this leads to
much more IO than is needed. If you couple this with the fact that EBS
is not a single spinning disk, and that a request is made on the
logical EBS volume, not physical underlying disk you are going to see
a larger number of iops having to be performed which will slow down
the IO operations from the host OS perspective.
Since each OS/Distro/AMI might set a different read-ahead value it is
best to test for yourself, but many people have seen much improvement
in latency, and reducing IO, by decreasing the read-ahead. This is
true on all disks, but seems especially true with EBS (RAID).
I don't have any numbers to share if that is what you are asking.
> --
> You received this message because you are subscribed to the Google Groups
> "mongodb-user" group.