Fix Redit changing IDs
[gofetch.git] / test / expected / LWN / 0000763603.header
... / ...
CommitLineData
10[$] Measuring (and fixing) I/O-controller throughput loss null/LWN/0000763603 70\r
2i [Kernel] Aug 29, 2018 21:20 UTC (Wed) (corbet)\r
3i\r
4i Many services, from web hosting and video streaming to cloud\r
5i storage, need to move data to and from storage. They also\r
6i often require that each per-client I/O flow be guaranteed a\r
7i non-zero amount of bandwidth and a bounded latency. An\r
8i expensive way to provide these guarantees is to over-provision\r
9i storage resources, keeping each resource underutilized, and\r
10i thus have plenty of bandwidth available for the few I/O flows\r
11i dispatched to each medium. Alternatively one can use an I/O\r
12i controller. Linux provides two mechanisms designed to throttle\r
13i some I/O streams to allow others to meet their bandwidth and\r
14i latency requirements. These mechanisms work, but they come at\r
15i a cost: a loss of as much as 80% of total available I/O\r
16i bandwidth. I have run some tests to demonstrate this problem;\r
17i some upcoming improvements to the [1]bfq I/O scheduler promise\r
18i to improve the situation considerably.\r
19i \r
20i \r
21i \r
22i [1] https://lwn.net/Articles/601799/\r
23i\r