In this topic are listed the fundamental rules that should be considered before integrating the DDP into any network.
None of these rules should be neglected as it will directly affect the performance and stability of the overall system.
Unless our team explicitly gives you specific indications, there should not be any exceptions.
Network rules on the DDP side:
- Always disable flow control for 10 NICs going to a switch, always enable flow control for 1G NICs in general and 10G NICs directly going to client machines.
- Set MTU 9000 for 10 NICs directly going to client machines, set it to 1500 in all other cases. The MTU value on one DDP port must always correspond to the value set on the client machines side.
Network rules on the switch:
- Always disable flow control for 10 NICs going to a DDP, enable it for all other cases.
- Set the MTU setting to either 1500 or 9000.
Letting the MTU 9000 setting for all the switch ports is actually not a problem even when the client machines and the DDP are set to 1500.
The reason is that the switch will never send more than 1500 bytes packets if neither the DDP nor the client machines are set to MTU 9000.
- Full duplex setting must be enabled
Network rules on the client machines:
- Always enable flow control.
- Set MTU 9000 for 10 NICs directly going to a DDP, set it to 1500 in all other cases. The MTU value on the client machines ports must always correspond to the value set on the DDP port.
- Full duplex setting must be enabled.
- Any network power saving feature must be turned off
Again, MTU values must be consistent between the DDP and client machines ports. A very simple way to test if a remote machine port MTU is set to 9000 is to ping it as per the command below (replace the IP address with the remote machine one):
ping -f -L 9000 172.16.5.100
This command does not work, it means that either your switch or your remote machine MTU is not configured properly.
If you can get a reply from the remote client, it means that both your switch and your host have their MTU properly configured.
For the flow control setting on the 10G ports of both the DDP and switch, there are actually two options:
- Switch it off and accept packet loss to occur
This first option is the one we are mainly recommending as it is the easiest and quickest one, but some applications that do not pre-buffer the media like FCP are sensitive to packet loss
- Switch it on and configure traffic control to prevent the switch buffer to fill up, causing head of line blocking
This second option can be chosen to avoid packet loss and get a higher effective throughput on the 10G uplink, however configuring and applying the traffic control script takes more time as it must be adapted depending on the network infrastructure.
When severals switches are getting nested, the same rules apply. So when a DDP port is connected with a 10G cable to a switch, that is itself connected to a switch with a 10G cable, the flow control setting should be – either turned off on both switches – either turned on on both switches with a traffic control script running on the DDP.
Please note that the graphical user interface of OSX network preferences can be misleading when modifying the MTU setting and might not reflect the actual value, also after a client machine reboot the setting might be cleared if the network interface does not support properly jumbo frames.
For that reason it is highly recommended to check the actual value using the ifconfig command.