test/expected/LWN/0000763603.header

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