backend network configuration
shulemmosko
33 Posts
December 17, 2024, 9:59 pmQuote from shulemmosko on December 17, 2024, 9:59 pmHi, we currently run a 5 node cluster, each node has 2x 40Gb. we have 2x 40Gb switches stacked. but currently we are only using one port on each node connected to one switch the second port on the nodes and the second switch is not doing anything.
i would like to setup some failover and also getting 80Gb throughput for the backend.
What would be my best approach? to bond both 40Gb ports and use one ip and create a LAG/Port-Channel on the switch? or to use two backend ip's? or any other approach?
Thanks!
Shulem
Hi, we currently run a 5 node cluster, each node has 2x 40Gb. we have 2x 40Gb switches stacked. but currently we are only using one port on each node connected to one switch the second port on the nodes and the second switch is not doing anything.
i would like to setup some failover and also getting 80Gb throughput for the backend.
What would be my best approach? to bond both 40Gb ports and use one ip and create a LAG/Port-Channel on the switch? or to use two backend ip's? or any other approach?
Thanks!
Shulem
f.cuseo
66 Posts
December 18, 2024, 8:19 amQuote from f.cuseo on December 18, 2024, 8:19 amI have 4 x 10Gbit in a LACP bond, and I use a different vlan for every service (one for management, one for ceph, one for s3, one for nfs).
Of course, remember that with bonded ports your traffic will be redirected on different physical ports based to hashing algoritm.
I have 4 x 10Gbit in a LACP bond, and I use a different vlan for every service (one for management, one for ceph, one for s3, one for nfs).
Of course, remember that with bonded ports your traffic will be redirected on different physical ports based to hashing algoritm.
shulemmosko
33 Posts
December 18, 2024, 9:41 pmQuote from shulemmosko on December 18, 2024, 9:41 pmThanks!
but this is an existing production cluster, im going to have to do one node at a time.
is there any petasan configs i need to update? since currently the backend is using eth4 and once i bond it with eth5 ill probably need to update petasan config to use the bond name?
please advise!
Thanks much!!
Thanks!
but this is an existing production cluster, im going to have to do one node at a time.
is there any petasan configs i need to update? since currently the backend is using eth4 and once i bond it with eth5 ill probably need to update petasan config to use the bond name?
please advise!
Thanks much!!
admin
2,929 Posts
December 19, 2024, 9:32 pmQuote from admin on December 19, 2024, 9:32 pmyou need to update
/opt/petasan/config/cluster_info.json
you need to update
/opt/petasan/config/cluster_info.json
backend network configuration
shulemmosko
33 Posts
Quote from shulemmosko on December 17, 2024, 9:59 pmHi, we currently run a 5 node cluster, each node has 2x 40Gb. we have 2x 40Gb switches stacked. but currently we are only using one port on each node connected to one switch the second port on the nodes and the second switch is not doing anything.
i would like to setup some failover and also getting 80Gb throughput for the backend.
What would be my best approach? to bond both 40Gb ports and use one ip and create a LAG/Port-Channel on the switch? or to use two backend ip's? or any other approach?
Thanks!
Shulem
Hi, we currently run a 5 node cluster, each node has 2x 40Gb. we have 2x 40Gb switches stacked. but currently we are only using one port on each node connected to one switch the second port on the nodes and the second switch is not doing anything.
i would like to setup some failover and also getting 80Gb throughput for the backend.
What would be my best approach? to bond both 40Gb ports and use one ip and create a LAG/Port-Channel on the switch? or to use two backend ip's? or any other approach?
Thanks!
Shulem
f.cuseo
66 Posts
Quote from f.cuseo on December 18, 2024, 8:19 amI have 4 x 10Gbit in a LACP bond, and I use a different vlan for every service (one for management, one for ceph, one for s3, one for nfs).
Of course, remember that with bonded ports your traffic will be redirected on different physical ports based to hashing algoritm.
I have 4 x 10Gbit in a LACP bond, and I use a different vlan for every service (one for management, one for ceph, one for s3, one for nfs).
Of course, remember that with bonded ports your traffic will be redirected on different physical ports based to hashing algoritm.
shulemmosko
33 Posts
Quote from shulemmosko on December 18, 2024, 9:41 pmThanks!
but this is an existing production cluster, im going to have to do one node at a time.
is there any petasan configs i need to update? since currently the backend is using eth4 and once i bond it with eth5 ill probably need to update petasan config to use the bond name?
please advise!
Thanks much!!
Thanks!
but this is an existing production cluster, im going to have to do one node at a time.
is there any petasan configs i need to update? since currently the backend is using eth4 and once i bond it with eth5 ill probably need to update petasan config to use the bond name?
please advise!
Thanks much!!
admin
2,929 Posts
Quote from admin on December 19, 2024, 9:32 pmyou need to update
/opt/petasan/config/cluster_info.json
you need to update
/opt/petasan/config/cluster_info.json