Monitor change and possible bottlenecks
Ste
125 Posts
February 12, 2021, 10:10 amQuote from Ste on February 12, 2021, 10:10 amGood morning, I'm going to install a new Petasan cluster and I would like to ask some clarifications about monitors:
- Once I setup the cluster and define which are the 3 monitors (3 of the 4 data nodes), then will I be able to replace one of them in the future (let's say add a new server and make this new one a monitor removing one from the original three) ?
- If this new monitor is not a high-end server, with less powerful CPU, less RAM and 1 Gb/s ethernet, can this create a bottleneck in the cluster, compared to the situation where all the monitors are also storage nodes (Xeon CPU, lots of RAM and 10 Gb/s ethernet) ?
- In general: would the choice of having a simple dedicated monitor (in total 1 monitor + 2 storage/monitor + 2 storage only) bring some advantage in terms of performance and reliability of the cluster (compared to 3 storage/monitor + 1 storage only), or it would have no effects at all, or even be worse ?
Thanks for any explaination and bye, S.
Good morning, I'm going to install a new Petasan cluster and I would like to ask some clarifications about monitors:
- Once I setup the cluster and define which are the 3 monitors (3 of the 4 data nodes), then will I be able to replace one of them in the future (let's say add a new server and make this new one a monitor removing one from the original three) ?
- If this new monitor is not a high-end server, with less powerful CPU, less RAM and 1 Gb/s ethernet, can this create a bottleneck in the cluster, compared to the situation where all the monitors are also storage nodes (Xeon CPU, lots of RAM and 10 Gb/s ethernet) ?
- In general: would the choice of having a simple dedicated monitor (in total 1 monitor + 2 storage/monitor + 2 storage only) bring some advantage in terms of performance and reliability of the cluster (compared to 3 storage/monitor + 1 storage only), or it would have no effects at all, or even be worse ?
Thanks for any explaination and bye, S.
Last edited on February 12, 2021, 10:11 am by Ste · #1
admin
2,930 Posts
February 12, 2021, 7:03 pmQuote from admin on February 12, 2021, 7:03 pmYou can replace the management nodes, when you deploy a node choose "Replace Management Node". It needs to have the same hostname and ip address.
You can have these nodes without storage, but they should not be underpowered, beside the Ceph monitors the management nodes also acts as Ceph mgr, mds, Consul servers, Gluster servers, Stats servers. Ceph monitors also can get quite busy in case a large cluster is recovering as they need to issue cluster maps.
You can replace the management nodes, when you deploy a node choose "Replace Management Node". It needs to have the same hostname and ip address.
You can have these nodes without storage, but they should not be underpowered, beside the Ceph monitors the management nodes also acts as Ceph mgr, mds, Consul servers, Gluster servers, Stats servers. Ceph monitors also can get quite busy in case a large cluster is recovering as they need to issue cluster maps.
Monitor change and possible bottlenecks
Ste
125 Posts
Quote from Ste on February 12, 2021, 10:10 amGood morning, I'm going to install a new Petasan cluster and I would like to ask some clarifications about monitors:
- Once I setup the cluster and define which are the 3 monitors (3 of the 4 data nodes), then will I be able to replace one of them in the future (let's say add a new server and make this new one a monitor removing one from the original three) ?
- If this new monitor is not a high-end server, with less powerful CPU, less RAM and 1 Gb/s ethernet, can this create a bottleneck in the cluster, compared to the situation where all the monitors are also storage nodes (Xeon CPU, lots of RAM and 10 Gb/s ethernet) ?
- In general: would the choice of having a simple dedicated monitor (in total 1 monitor + 2 storage/monitor + 2 storage only) bring some advantage in terms of performance and reliability of the cluster (compared to 3 storage/monitor + 1 storage only), or it would have no effects at all, or even be worse ?
Thanks for any explaination and bye, S.
Good morning, I'm going to install a new Petasan cluster and I would like to ask some clarifications about monitors:
- Once I setup the cluster and define which are the 3 monitors (3 of the 4 data nodes), then will I be able to replace one of them in the future (let's say add a new server and make this new one a monitor removing one from the original three) ?
- If this new monitor is not a high-end server, with less powerful CPU, less RAM and 1 Gb/s ethernet, can this create a bottleneck in the cluster, compared to the situation where all the monitors are also storage nodes (Xeon CPU, lots of RAM and 10 Gb/s ethernet) ?
- In general: would the choice of having a simple dedicated monitor (in total 1 monitor + 2 storage/monitor + 2 storage only) bring some advantage in terms of performance and reliability of the cluster (compared to 3 storage/monitor + 1 storage only), or it would have no effects at all, or even be worse ?
Thanks for any explaination and bye, S.
admin
2,930 Posts
Quote from admin on February 12, 2021, 7:03 pmYou can replace the management nodes, when you deploy a node choose "Replace Management Node". It needs to have the same hostname and ip address.
You can have these nodes without storage, but they should not be underpowered, beside the Ceph monitors the management nodes also acts as Ceph mgr, mds, Consul servers, Gluster servers, Stats servers. Ceph monitors also can get quite busy in case a large cluster is recovering as they need to issue cluster maps.
You can replace the management nodes, when you deploy a node choose "Replace Management Node". It needs to have the same hostname and ip address.
You can have these nodes without storage, but they should not be underpowered, beside the Ceph monitors the management nodes also acts as Ceph mgr, mds, Consul servers, Gluster servers, Stats servers. Ceph monitors also can get quite busy in case a large cluster is recovering as they need to issue cluster maps.