Project

General

Profile

Actions

Feature #4239

open

Automatically determine routing data based on host routing data (entirely deprecate public/cluster explicit specification)

Added by Michael Evans about 11 years ago. Updated about 11 years ago.

Status:
New
Priority:
Low
Assignee:
-
Category:
-
Target version:
-
% Done:

0%

Source:
Community (user)
Tags:
Backport:
Reviewed:
Affected Versions:
Pull request ID:

Description

IPv6 has interesting features, such as the expectation that hosts have multiple addresses per interface (local link, public IPs, private IPs).

It might be more useful to require only that all daemons can talk to other daemons over at least one route, but not necessarily that any single daemon need communicate with each other daemon via the same route. Implicit in that is that not all daemons need be on a single network. (A fictional and easily constructed topology would be three servers connected via crossover cable in a (3 cable total) mesh topology, with each physical cable assigned it's own network segment. This might be desirable in case they are mistakenly wired to the same network as routing and diagnostics would still function.)

Each daemon would keep track of three sets of routing metrics. The first would be the host system's local routing metric table. The second would reflect the current performance expectations of each link, while the third would be the score over some slightly recent interval. Every so often the third would be updated to the mons.

When establishing communication the common path between any two daemons with the lowest score would then be the winner; such a score might consider the likely bias of information flow as an additional modifier to each metric.

No data to display

Actions

Also available in: Atom PDF