Hi, On 04.04.2020 14:16, Volodymyr Litovka wrote:
while for more scaled systems, where most of VMs will reside on diskless nodes, this will have a notable performance impact
For example, in Openstack, where Nova knows nothing about Cinder's volumes placement, there are no ways (at least - simple) to reside VM on same node with volume. So, e.g. in the project with 5+ nodes and two replicas, probability of residing VM on "diskless" node becomes very high.
The question is - whether DRBD supports multicast in order to generate a single flow from originating node? This will allow to dramatically improve performance of scaled system and will allow to build more scaled systems.
This change however, can significantly impact the source code and will require much efforts. To avoid this, I'd propose to add additional layer like proxy, which will: - reside on every satellite - receive data over unicast ** thus, drbd code will get minimal changes (now - it sends unicast data to 1+ neighbors, after changes - it will send the same unicast to single neighbor) ** to minimize delay - use local sockets - resend it over multicast - but manage control traffic (e.g. acknowledgments from remote peers) over unicast Thank you. -- Volodymyr Litovka "Vision without Execution is Hallucination." -- Thomas Edison _______________________________________________ Star us on GITHUB: https://github.com/LINBIT drbd-user mailing list [email protected] https://lists.linbit.com/mailman/listinfo/drbd-user
