Netgate is offering COVID-19 aid for pfSense software users, learn more.
Troubleshooting Low Interface Throughput¶
In situations where the firewall is not transferring as much data as desired. There are many potential causes for this condition, most of which are listed here along with possible resolutions
The first thing to check is that the hardware is capable of pushing the expected amount of traffic. In some cases this is more obvious, such as a newer multi-core server being unable to transfer small amounts of packets, or an older firewall not being able to transfer high loads. Other cases are more subtle and require some testing and verification.
The most obvious test is to watch the firewall CPU load while transferring data. This can be observed from Diagnostics > System Activity or from the shell by running:
If an IRQ process for a network card is using a significant amount of CPU on a core, then either the hardware is being fully (or over) utilized, or the driver may need adjustments to work as expected. If the firewall is not under any stress whatsoever while transferring data, the problem likely lies elsewhere.
If the amount of “System” CPU is high and the amount of interrupts is low, the problem may be in the amount of packet processing happening in pf or being used for encryption. If pf is pushing the CPU as high as it can, it may require a faster CPU.
If the CPU is being used for encryption, a faster cipher may be chosen, or in some cases a cryptographic accelerator may be utilized.
Hardware/Driver Tuning Required¶
If a CPU core is fully utilized by interrupts, the network card driver may need tuning. Most of these tweaks are covered on Tuning and Troubleshooting Network Cards. Some cards, such as igb, are able to use more queues for processing packets which will spread the load across multiple cores and result in higher throughput, but not every workload is helped by these options, so less queues may also help.
Another item to check is under System > Advanced on the Networking tab. Ensure that the boxes are checked for Disable hardware TCP segmentation offload and Disable hardware large receive offload. If they are already checked, try toggling Disable hardware checksum offload. If no difference is observed, toggle it back.
A duplex mismatch is also possible, though this is more common on circuits 100Mbit/s or less. Some providers are stuck in the stone age and still insist on hard-coding ports on CPEs such as fiber converters at 100Mbit/s full-duplex. If the CPE is hard-coded but the firewall is not, it would show as using 100Mbit/s half-duplex on Status > Interfaces. The duplex mismatch will lead to interface errors, collisions, and low throughput. Setting the speed and duplex is covered on Forcing Interface Speed or Duplex Settings.
If the traffic shaping wizard was run previously before an increase in upstream bandwidth, the old limits may still be in effect. Visit Firewall > Traffic Shaper and check the root interface queues, and qInternet queues to ensure that any listed interface bandwidths are appropriately specified and current.
Also check the Limiters tab under the traffic shaper settings, verify that any configured limiters are set for appropriate speeds. Limiters may also need increased queue lengths to handle higher throughput volumes.
Issues with upload speed frequently end up being issues with the MTU. If the MTU on pfSense® software (default 1500), is higher than the MTU of the upstream link, it can result in packets being fragmented, lost, or otherwise mishandled. Setting MSS clamping on the WANs or changing the MTU of the interface may help.
VPN + MTU Issues¶
Similar to the above, if large packets or high-throughput seems to break over a
VPN, enable MSS Clamping for VPN Networks under VPN > IPsec, Advanced
Settings tab. The default value for the option is
1400, but try lower
values such as
There could also be issues between the WAN and the Modem/CPE. It could be a cable, or a quirk in how the two interfaces talk to each other. Place a small switch between the firewall and the Modem/CPE as a test.
The slowness may not be from any cause on the firewall. It could be the client itself or how it connects. Testing a 100Mbit/s WAN over 802.11g wireless, for example, would never show full speed. Testing a 300Mbit/s WAN from a 100Mbit/s LAN connection would likewise not be a valid test.
Ensure the client is connected to the firewall through a connection at least as fast as the WAN supports.
If every other factor has been eliminated, test the modem without the firewall involved. If the speed is still low, it may be the ISP to blame, or the Modem/CPE.