High number of interrupts with Jetway NF9HG-2930 (4 x Intel i211AT)
-
I'm currently testing a Jetway NF9HG-2930 (4 x Intel i211AT). I'm noticing it has a lot more interrupt calls than other boards I have used. On the RRD graphs, I'm seeing 5-10 on the one minute resolution and up to 20 on the 5 minute averages.
The only tuning I've done was to set kern.ipc.nmbclusters to 1000000 per document (https://doc.pfsense.org/index.php/Tuning_and_Troubleshooting_Network_Cards). I'm not sure if this is a problem, or if there are other tuning parameters I should set.
The firewall seems to be performing well, but with the holiday, it's not being pushed as hard as it typically would.
Thanks in advance
-
The only tuning I've done was to set kern.ipc.nmbclusters to 1000000 per document (https://doc.pfsense.org/index.php/Tuning_and_Troubleshooting_Network_Cards).
Pending on your RAM size this would be a fine thing, each CPU core will create for each LAN
port one queue! So if you have 4 CPU cores and 4 LAN Ports your box must handle 16 queues.
If you have enough RAM it would be fine!I'm not sure if this is a problem, or if there are other tuning parameters I should set.
For the mSATA or SSD you could enable TRIM support in pfSense.
If you are using Squid you could also high up the default (256 MB) of the RAM size Squid is using
to speed up the proxy.You should enable PowerD (hi adaptive) that even the right CPU frequency is used pending on the
needed power called by action or the enabled services. -
I have 4 Gb RAM in the machine (showing 17% used), non-SSD disk, and not running Squid. Does this all sound about right?
-
Does this all sound about right?
Yep, but what is about the PowerD (hi adaptive) option?
-
Could you please explain what PowerD is? I'm not familiar with it.
-
I did find the PowerD setting in Advanced Misc. It is currently off. Should it on and set to HiAdaptive?
-
Could you please explain what PowerD is? I'm not familiar with it.
It is regulating the CPU frequencies form the lower bottom to the highest top available.
Otherwise your CPU is perhaps (not even) locked to 800MHz and if more power is needed
it would be not able to serve! Otherwise it could be that your CPU is even and only running on
its maximum frequency and this is also not what you will need or have.PowerD:
- minimum
even the minimum cpu frequency is used - maximum
even the maximum cpu frequency is used - adaptive
mixed between minimum and maximum but with a viewing eye on power saving - hi adaptive
mixed between minimum and maximum but with a viewing eye on throughput
I did find the PowerD setting in Advanced Misc. It is currently off. Should it on and set to HiAdaptive?
Yes for sure because your CPU frequency is from 2,16GHz to 2,4GHz turbo boost mode and to get it right working
PowerD is there, not alon for that, but also! Otherwise, likes often seen at PC Engines APU boards, you
will perhaps your CPU even run on 600MHz or 800Mhz and if more power is really needed it could not
be delivered by the CPU. With enabled PowerD the clock frequency is going from its minimum to its
maximum likes the power is needed or called by the entire pfSense system. - minimum
-
What does your CPU RRD look like? How much traffic through the system (throughput graph to correlate to CPU would help)?
-
CPU almost never is getting over 20 on the graph, typically less than 10. As the CPU spikes, so does the throughput graph. When the throughput maxes out our connection, the CPU still has lots more capacity.
-
Attached are the two graphs.
-
Looks normal, it's only an average of 2, and moves in relation to the throughput and never gets all that high.
-
Thanks for looking. I guess it's just a different Ethernet adapter. My first pfSense hardware used Realtek adapters and when I saw any interrupt usage, the network would stall. Hence, I'm a little scared when I see any usage. Most of my other pfSense boxes do not show any interrupt usage.
-
Most of my other pfSense boxes do not show any interrupt usage.
If the entire rest in pfSense is running smooth and liquid I would not care about this "behavior".
-
They'll all show some interrupt usage, that's part of how computers work. If your hardware's significantly oversized, especially CPU, relative to the amount of traffic you're pushing, it's probably a fraction of a percent. It's perfectly normal to have more interrupt load than that though, including well beyond what you're seeing if you start pushing a lot more traffic.
The interrupt load itself wasn't a problem with your Realtek NICs. Some of those NICs just fall apart under load, and the interrupt load increasing is just what happens when your NICs are under more load.