Skip to content

RFC dm-mpath: service-time oriented dynamic load balancer

Kiyoshi Ueda dm-devel 1 version New

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest0 patches 0 repliesNew

Patches in v1 No applies recorded yet

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [RFC PATCH 0/0] dm-mpath: service-time oriented dynamic load balancer

Hi,

The following patches add a service-time oriented dynamic load balancer, dm-service-time, which selects a path to complete the incoming I/O with the shortest time.

While the dm-queue-length path selector posted in the other thread is ready for inclusion, I would rather like to hear comments from other dm-multipath users and developers for dm-service-time before pushing it to mainline.

dm-service-time stores the recent throughput information for each path. Service time for a incoming I/O is estimated by:
("the total size of in-flight I/Os on the path" +
"the size of the incoming I/O") / "the recent throughput"

There could be simpler path selection methods:
- Size-based path selection.
i.e. just use the total size of in-flight I/Os on the path. It is simpler and works fine if each path has same bandwidth, is evenly loaded and has no disturbance from external system.
- Use user-specified performance value for each path. i.e. instead of calculating the throughput in kernel, the value is given from userspace.
It works fine if the bandwidth of paths are asymmetric but there is no disturbance from external system.

dm-service-time is more adaptive than the above 2 methods. However, if the simpler methods are preferred in some cases, I can split the dm-service-time into the pieces to allow the simpler approaches.

This patch-set can be applied on top of 2.6.29-rc2 + dm-queue-length patches(*).

(*) https://www.redhat.com/archives/dm-devel/2009-January/msg00183.html

Summary of the patch-set:
1/2: dm-mpath: interface change for preparation
2/2: dm-mpath: add service-time oriented dynamic load balancer

Thanks,
Kiyoshi Ueda