I have 10 proxmox VM hosts and a shared storage.
Each proxmox host has 2x gigabit ports.
The shared storage has 6x gigabit ports.
I currently have each node set to do LACP with the switch using 2 ports. The same goes for the storage, with 6 ports.
Everything works fine but, offcourse, my VM's disk performance is limited to 1 gigabit. Since bonding, LACP, won't strip traffic across multiple ports.
Now I want higher throughput inside my VM's (disk performance) and I'm looking at the options that I have.
The best/easiest option is upgrading to 10G. But that's not possible right now (too expensive).
The other options I can think of are:
1. Using the balance-rr mode of the bonding driver. It'll stripe traffic across multiple interfaces. I could setup vlans and have traffic striped across it, should work according to this document: http://louwrentius.com/achieving-450-mbs-network-file-transfers-using-linux-bonding.html
or
2. Switch from NFS to iSCSI, more specifically iSCSI multipath.
I'm not a fan of iSCSI because I had bad performance tests a while ago, when deciding between NFS or iSCSI. At that time I had considerably better performance over NFS when there was random i/o in the VM. Also throughput was better.. But maybe I did something wrong ? ...
How do you guys increase throughput when using shared storage?
Or are you guys all rich and buy 10g stuff?
Each proxmox host has 2x gigabit ports.
The shared storage has 6x gigabit ports.
I currently have each node set to do LACP with the switch using 2 ports. The same goes for the storage, with 6 ports.
Everything works fine but, offcourse, my VM's disk performance is limited to 1 gigabit. Since bonding, LACP, won't strip traffic across multiple ports.
Now I want higher throughput inside my VM's (disk performance) and I'm looking at the options that I have.
The best/easiest option is upgrading to 10G. But that's not possible right now (too expensive).
The other options I can think of are:
1. Using the balance-rr mode of the bonding driver. It'll stripe traffic across multiple interfaces. I could setup vlans and have traffic striped across it, should work according to this document: http://louwrentius.com/achieving-450-mbs-network-file-transfers-using-linux-bonding.html
or
2. Switch from NFS to iSCSI, more specifically iSCSI multipath.
I'm not a fan of iSCSI because I had bad performance tests a while ago, when deciding between NFS or iSCSI. At that time I had considerably better performance over NFS when there was random i/o in the VM. Also throughput was better.. But maybe I did something wrong ? ...
How do you guys increase throughput when using shared storage?
Or are you guys all rich and buy 10g stuff?