Tom Lane wrote:
> Scott Carey <scott@richrelevance.com> writes:
>> Which brings this back around to the point I care the most about:
>> I/O per second will diminish as the most common database performance limiting factor in Postgres 8.4's lifetime, and
becomealmost irrelevant in 8.5's.
>> Becoming more CPU efficient will become very important, and for some, already is. The community needs to be
proactiveon this front.
>> This turns a lot of old assumptions on their head, from the database down through the OS and filesystem. We're
boundto run into many surprises due to this major shift in something that has had its performance characteristics taken
forgranted for decades.
>
> Hmm ... I wonder whether this means that the current work on
> parallelizing I/O (the posix_fadvise patch in particular) is a dead
> end. Because what that is basically going to do is expend more CPU
> to improve I/O efficiency. If you believe this thesis then that's
> not the road we want to go down.
>
I imagine the larger postgres installations will still benefit from
this patch - because I imagine they will stay on hard disks for
quite some time; simply because the cost of 70TB of disks seems like
it'll be lower than RAM for at least the intermediate term.
I imagine the smaller postgres installations will also still benefit
from this patch - because my postgres installations with the most
painful I/O bottlenecks are small virtual machines without dedicated
drives where I/O (I guess emulated by the virtual machine software)
is very painful.
Perhaps there's a mid-sized system that won't benefit from fadvise()
in the intermediate term -- where the size of the database is about
the same size as a cost-effective flash drive -- but I don't have
any databases in that range now.