a solution to the severe throughput loss caused by I/O-bandwidth management

Hi,
this is to inform you guys about the severe throughput loss caused by the techniques currently used to manage I/O bandwidth, and about a new solution to address this issue.

To isolate containers when also I/O is involved, it is necessary to guarantee a minimum I/O bandwidth to each container. Disappointingly, the techniques used to guarantee I/O bandwidths entail dramatic throughput losses: up to 80-90% of the storage throughput. Details, e.g., in this recent post of mine on Linaro’s blog:
http://ow.ly/vsrW50mBAGl

On the bright side, I’ve devised a solution that reduces this loss to just 10% (I’m a developer of the BFQ I/O scheduler [1, 2]). This result too is covered in the above post.

I’ll be happy to answer any question I can answer :slight_smile:

Thanks,
Paolo

[1] https://www.kernel.org/doc/Documentation/block/bfq-iosched.txt
[2] https://algo.ing.unimo.it/people/paolo/disk_sched/results.php