0[$] Measuring (and fixing) I/O-controller throughput loss	null/LWN/0000763603		70
i  [Kernel] Aug 29, 2018 21:20 UTC (Wed) (corbet)
i
i    Many  services,  from  web hosting and video streaming to cloud
i    storage,  need  to  move  data  to  and from storage. They also
i    often  require  that  each  per-client I/O flow be guaranteed a
i    non-zero   amount  of  bandwidth  and  a  bounded  latency.  An
i    expensive  way to provide these guarantees is to over-provision
i    storage  resources,  keeping  each  resource underutilized, and
i    thus  have  plenty of bandwidth available for the few I/O flows
i    dispatched  to  each  medium.  Alternatively one can use an I/O
i    controller.  Linux provides two mechanisms designed to throttle
i    some  I/O  streams  to allow others to meet their bandwidth and
i    latency  requirements.  These mechanisms work, but they come at
i    a  cost:  a  loss  of  as  much  as  80% of total available I/O
i    bandwidth.  I  have run some tests to demonstrate this problem;
i    some  upcoming improvements to the [1]bfq I/O scheduler promise
i    to improve the situation considerably.
i    
i    
i    
i    [1] https://lwn.net/Articles/601799/
i