speedup live migration

[PointPubMedia]

Hey folks..

We are having a 6 nodes cluster right now, using a 4ports NIC using LACP for VM traffics and also a virtual interface for the corosync traffic and another 4 ports NIC for the iSCSI traffic.

Am I right to say that the migration traffic is using the "LACP" traffic and not the corosync interface?

How can we speedup drastically the speed of live migration ?

 

[wolfgang]

Hi,

How can we speedup drastically the speed of live migration ?

with a faster nic.

LACP is not able to increase the bandwidth for single connections.
If your switch would support RoundRobin this would increase the bandwidth.
But do this only if the switch supports it. If, not you get problems with the packed order.

 

[spirit]

It could be possible with qemu multithreaded live migration. (Multiple connections, parallel ). But its not implemented currently. Maybe Can you full a request to proxmox bugzilla ?

 

[spirit]

It could be done with:

migrate_set_capability multifd on
migrate_set_parameter multifd-channels 4 (number of links)

on source guest.

(I just need to check how it's work with proxmox ssh tunnel, or with socat with ssh is disabled)

 

[adirburke]

It could be done with:

migrate_set_capability multifd on
migrate_set_parameter multifd-channels 4 (number of links)

on source guest.

(I just need to check how it's work with proxmox ssh tunnel, or with socat with ssh is disabled)

Hey man, sorry to re-vive a dead post, where do you put those settings?

 

[fiona]

Hi,

Hey man, sorry to re-vive a dead post, where do you put those settings?

this is currently not implemented and requires changes to the Proxmox VE source code. I've evaluated multifd migration a while ago, and it worked fine, but requires a bit restructuring of the current migration code. In particular, we'd need to use -incoming defer when starting the target instance and set the multifd parameters on the source and target before starting the actual migration.