The problem

Sites which are far away from each other pose a significant problem for HA solutions because the communication channels between them are both expensive and unreliable. Apart from needing a different protocol and product for communication, the biggest issue is the impossibility of deploying fencing. All supported stonith devices are accessible either via network interface or serial interface or similar, but in order to operate them from the remote site access to the local network is required and that may not be available due to the split brain condition. It is unsuitable to use the same communication channels for both cluster communication and fencing.

The solution

booth coordinates and serves geographically dispersed sites which are capable of running the same service or services. However, it cannot guarantee that a site which is not reachable has released resources. Another, possibly only intermittently available but affordable, channel may be used to access the unreachable site and perform some operation which may establish the definite state of shared resources. Some possibilities include plain modem or even SMS text messaging.

The implementation

All building blocks are readily available: PPP for modem communication or some SMS sending program (still need authentication for the latter). Some existing protocol may be adopted for the messaging between sites. What probably needs to be implemented is a program to handle messages. It may or may not be part of booth.

Other details TBD.

Looking for mad skills in:

Nothing? Add some keywords!

This project is part of:

Hack Week 12

Activity Show All

  • over 2 years ago: dmuhamedagic originated Slow and OOB channel for intercluster communication
  • Comments