Slow performance

Hello,

I’m running a virtualized infrastructure with Proxmox. It’s equipped with Dell servers and a 10Gbps network interface for DRBD synchronization. HDD SAS on PERC.

There’s one VM on each Proxmox server, with a primary node and a secondary node. Protocol C is used. The 10Gbps interfaces are Proxmox Bridges.

When the primary node isn’t connected to secondary node, disk performance is acceptable (fio):
r=2512MiB/s,w=2517MiB/s

As soon as the secondary node is connected to the interface, disk performance drops dramatically:
r=84MiB/s,w=52MiB/s

Do you have any ideas?

Thanks.

DRBD can really only operate as fast as the slowest bottle-neck. In cases like this there is likely a bottle-neck somewhere.

How fast is the disk in the peer? Can you test its performance while also disconnected?

Can you confirm the network is running at 10gb (~1100/MiB)? What is the network latency between the hosts like (ping RTT)?

On the peer, it’s a 6Gb/s SAS HDD on a PERC card.

Disconnected performance is around ~3000/MiB with BBU caching.

I can confirm that I’m achieving 10Gbps via iperf3. And ping <1ms.

The network connection is peer-to-peer with no intermediate equipment.