FQ is a fair-queuing policy, primarily designed to balance the I/O between different processes taking into account the average latency of each request and the number of requests. It additionally avoids read starvation by giving higher priority to reads whenever they are available. It essentially should improve interactivity when I/O intensive (write intensive in particular) processes are running, like hammer rebalance, etc.
There are no known plans to add more scheduling policies, but I for one would very much welcome it. Writing new policies for the 'dsched' framework is really simple and, while fq does a reasonably good job, there is room for a lot of improvement in the form of further scheduling policies. I also added writing dsched policies as a gsoc project, and I would very much welcome someone doing that as a project. Hope that helps, Alex Hornung On 6 March 2011 17:48, Pierre Abbat <p...@phma.optus.nu> wrote: > I found a bunch of articles on disk scheduling in Wikipedia and checked what's > available in DF: > > # sysctl dsched.policies > dsched.policies: noop fq > > What does "fq" do? Are there plans to add any more? > > Pierre > -- > Jews use a lunisolar calendar; Muslims use a solely lunar calendar. >
