Just purchased a Netgate 8200 -- having a few issues.

stephenw10

That ALTQ setting only does anything to hn NICs (hyper-v or Azure). You can disable it.

You might try enabling (by unchecking) the off loading for LRO and TSO. In some situations that can help quite a bit.

inferno480

@stephenw10 No luck :( Unchecked all three boxes, Halted & pwr-cycled the 8200... booted it back up and I basically get the same results (consistently inconsistent results ranging from 3Gb/s-6Gb/s up and/or down, w/2-3 cores maxing out at 100%). I also tried to disable firewalling completely, and my testing was still around 3.5-3.6Gb/s w/iperf running on the firewall (same as before, but as you said -- not that great of a test).

Unfortunately I didn't really have an easy way to "test through" the 8200 w/NAT disabled, though... the Speedtest app from the Windows Store as well as the speedtest.net website seem IPv4-dependant. I'll see if I can get a pure IPv6 speedtest from somewhere.

Any other ideas? Any intel NIC chipset tunables I should maybe try tweaking? I would be happy if I could consistently/reliably achieve half of what the advertised specs are at this point.

stephenw10

Check the boot logs. Make sure you are seeing 8 queues per NIC.

Check the output of: vmstat -i

tman222

Hi @inferno480 -

Here's some more information on network tuning:

I agree with @stephenw10 that the first thing to check is to make sure that you have the right amount of RX and TX queues configured for your 8 core system. If that is the case, there are other settings you could try to tune as well: The first link above is more comprehensive FreeBSD network tuning guide, the second link has hardware tuning recommendations from the Netgate documentation (check out the Intel ix(4) Cards section). The third link is the manual entry for iflib which is a driver framework FreeSBD uses for network interface drivers. You can tune settings there by adding entries into your loader.conf.local file. The general syntax is dev.<interface name>.<interface number>.iflib.<tunable>=<value>. For example, dev.ix.0.iflib.override_nrxqs="8". Increasing the number of RX and TX descriptors might help, as well as increasing the rx_budget. I have not tried changing the separate_txrx parameter, but it might help give a better packet flow distribution across the cores.

Hope this helps.

inferno480

@tman222 Thanks for the tips, unfortunately still having the same issue of consistently inconsistent speeds that aren't even half of what's advertised. :(

Prior to making any of these changes, my speed tests (in Mbits/s) were:

6485 down, 3689 up
5808 down, 5027 up
3098 down, 5017 up
3969 down, 4638 up

I made the two recommended ix driver tweaks from the Netgate docs, and also the iflib changes. I did not touch any generic FreeBSD network tuning (yet) and had a question on the RX/TX Descriptors settings, but here is my /boot/loader.conf.local:

kern.ipc.nmbclusters="1000000"
kern.ipc.nmbjumbop="524288"
dev.ix.0.iflib.override_nrxqs="8"
dev.ix.0.iflib.override_ntxqs="8"
dev.ix.1.iflib.override_nrxqs="8"
dev.ix.1.iflib.override_ntxqs="8"
dev.ix.0.iflib.separate_txrx="1"
dev.ix.1.iflib.separate_txrx="1"
dev.ix.0.iflib.rx_budget="64"
dev.ix.1.iflib.rx_budget="64"

Wasn't quite sure what to adjust the RX/TX Descriptors to since I do not know the current values or what this particular chipset wants for syntax...

override_nrxds
Override the number of RX descriptors for each queue. The value
is a comma separated list of positive integers. Some drivers
only use a single value, but others may use more. These numbers
must be powers of two, and zero means to use the default.
Individual drivers may have additional restrictions on allowable
values. Defaults to all zeros.

override_ntxds
Override the number of TX descriptors for each queue. The value
is a comma separated list of positive integers. Some drivers
only use a single value, but others may use more. These numbers
must be powers of two, and zero means to use the default.
Individual drivers may have additional restrictions on allowable
values. Defaults to all zeros.

My speeds after the changes were:

5146 down, 3857 up
3167 down, 3989 up
5624 down, 4996 up
6786 down, 3843 up

I'm pretty sure it's using 8 queues, although I don't see any "TX Queues" in vmstat, hopefully that's normal:

from Status -> System Logs -> OS Boot:

(...)
ix0: <Intel(R) X553 N (SFP+)> mem 0x80400000-0x805fffff,0x80604000-0x80607fff at device 0.0 on pci9
ix0: Using 2048 TX descriptors and 2048 RX descriptors
ix0: Using 8 RX queues 8 TX queues
ix0: Using MSI-X interrupts with 9 vectors
ix0: allocated for 8 queues
ix0: allocated for 8 rx queues
ix0: Ethernet address: 90:ec:77:6c:af:7f
ix0: eTrack 0x8000084b PHY FW V65535
ix0: netmap queues/slots: TX 8/2048, RX 8/2048
ix1: <Intel(R) X553 N (SFP+)> mem 0x80200000-0x803fffff,0x80600000-0x80603fff at device 0.1 on pci9
ix1: Using 2048 TX descriptors and 2048 RX descriptors
ix1: Using 8 RX queues 8 TX queues
ix1: Using MSI-X interrupts with 9 vectors
ix1: allocated for 8 queues
ix1: allocated for 8 rx queues
ix1: Ethernet address: 90:ec:77:6c:af:80
ix1: eTrack 0x8000084b PHY FW V65535
ix1: netmap queues/slots: TX 8/2048, RX 8/2048
(...)

from 'vmstat -i':

[23.05.1-RELEASE][admin@pfSense.hadesgate.lan]/root: vmstat -i
interrupt total rate
irq16: sdhci_pci0+ 866 0
cpu0:timer 4280150 26
cpu1:timer 2495245 15
cpu2:timer 2470775 15
cpu3:timer 2570165 15
cpu4:timer 2623994 16
cpu5:timer 2690508 16
cpu6:timer 2600422 16
cpu7:timer 2582874 15
irq128: nvme0:admin 23 0
irq129: nvme0:io0 357532 2
irq130: nvme0:io1 361656 2
irq131: nvme0:io2 370062 2
irq132: nvme0:io3 366580 2
irq133: nvme0:io4 362204 2
irq134: nvme0:io5 367842 2
irq135: nvme0:io6 369601 2
irq136: nvme0:io7 367741 2
irq157: xhci0 1 0
irq158: ix0:rxq0 3153201 19
irq159: ix0:rxq1 3559443 21
irq160: ix0:rxq2 4185103 25
irq161: ix0:rxq3 4781562 29
irq162: ix0:rxq4 6998194 42
irq163: ix0:rxq5 5870912 35
irq164: ix0:rxq6 3420340 20
irq165: ix0:rxq7 4918691 29
irq166: ix0:aq 3 0
irq167: ix1:rxq0 5051730 30
irq168: ix1:rxq1 4969508 30
irq169: ix1:rxq2 5570319 33
irq170: ix1:rxq3 5975594 36
irq171: ix1:rxq4 7754642 46
irq172: ix1:rxq5 6481380 39
irq173: ix1:rxq6 5166430 31
irq174: ix1:rxq7 6701846 40
irq175: ix1:aq 6 0
irq176: ix2:rxq0 328394 2
irq177: ix2:rxq1 333122 2
irq178: ix2:rxq2 3866 0
irq179: ix2:rxq3 323267 2
irq180: ix2:rxq4 4147 0
irq181: ix2:rxq5 4011 0
irq182: ix2:rxq6 3790 0
irq183: ix2:rxq7 3769 0
irq184: ix2:aq 2 0
irq185: ix3:rxq0 325946 2
irq186: ix3:rxq1 3263 0
irq187: ix3:rxq2 53158 0
irq188: ix3:rxq3 322222 2
irq189: ix3:rxq4 2823 0
irq190: ix3:rxq5 2762 0
irq191: ix3:rxq6 324990 2
irq192: ix3:rxq7 3089 0
irq193: ix3:aq 2 0
Total 111839768 670

Should I be concerned about these nag messages in my OS Boot system logs?

iwi_bss: You need to read the LICENSE file in /usr/share/doc/legal/intel_iwi.LICENSE.
iwi_bss: If you agree with the license, set legal.intel_iwi.license_ack=1 in /boot/loader.conf.
module_register_init: MOD_LOAD (iwi_bss_fw, 0xffffffff8076b090, 0) error 1
iwi_ibss: You need to read the LICENSE file in /usr/share/doc/legal/intel_iwi.LICENSE.
iwi_ibss: If you agree with the license, set legal.intel_iwi.license_ack=1 in /boot/loader.conf.
module_register_init: MOD_LOAD (iwi_ibss_fw, 0xffffffff8076b140, 0) error 1
iwi_monitor: You need to read the LICENSE file in /usr/share/doc/legal/intel_iwi.LICENSE.
iwi_monitor: If you agree with the license, set legal.intel_iwi.license_ack=1 in /boot/loader.conf.
module_register_init: MOD_LOAD (iwi_monitor_fw, 0xffffffff8076b1f0, 0) error 1

inferno480

@inferno480 I should mention... core utilization behavior seems unchanged. Still just 2-3 cores maxing out during a speed test in either direction, with or without all the settings changes/tunings, with or without the NIC hardware offloading disabled.

tman222

@inferno480 - thanks for getting back to us.

Here are a few more things you could try:

The rx_budget budget value can be set a lot higher. I currently have mine set to the maximum, which is 65535.
The RX and TX descriptors can be increased all the way to up to 4096 I believe, and that may also be worth trying. The current values are 2048 for both RX/TX (you can see this in the boot logs you posted).
Have you tried a different speed test? For instance I know that with Netflix's speed test (Fast.com) you can tune the number of parallel connections in the Settings. Does increasing the number of parallel connections result in your connection bandwidth and CPU cores being more fully utilized?
Another tunable you may want to adjust is the hw.intr_storm_threshold value. Were you able to change that one? The value I would use there is 10000 as per the Netgate documentation: https://docs.netgate.com/pfsense/en/latest/hardware/tune.html#intel-ix-4-cards

Hope this helps.

tman222

@inferno480 - one thing I forgot to ask, what CPU was in your DIY build you used before?

inferno480

@tman222 So the latest suggestions did help... somewhat. Noticed the improvements after increasing descriptors to 4096. I did already have the interrupt threshold increased, but forgot to mention it as it was a System Tunable and not part of loader.conf.local.

5693.41 down, 3765.58 up
9197.79 down, 3792.50 up
8937.57 down, 3149.12 up
5580.54 down, 3987.94 up

I am now seeing download speed tests that are often in the more-acceptable range, but not very consistently. Now, 3-4 cores max-out instead of just 2-3. However, upload speeds are certainly still an issue and might possibly seem worse.

I have tried other speed test sites, but I can typically only get near-line rate (9Gb/s+) to my ISP's (Comcast). When using Fast.com, I get around 5.0Gb/s download and 1.5 Gb/s upload, even if I set it to a min & max of 8 streams (I've also tried 16). The weird thing with Fast.com is that initially speeds seem good (7-8Gb/s but then they "level off" to 5.0Gb/s over the duration of the test. From what I recall, these were about the speeds I would typically see to Fast.com on my old router... but the night & day difference is with my ISP's SpeedTest.net server using the Ookla SpeedTest.net app from the Windows Store.

So about that old server... massive overkill for a router. It was an ASUS P12R-E-10G-2T Motherboard with a Xeon E-2388G CPU with an old Intel X520-SR2 (dual SFP+ 10GbE NIC), 16GB RAM, running my primary 10G WAN and LAN ports.

This is my current loader.conf.local... I confirmed all the changes seem to be showing up in the OS Boot log.

legal.intel_iwi.license_ack="1"
kern.ipc.nmbclusters="1000000"
kern.ipc.nmbjumbop="524288"
dev.ix.0.iflib.override_nrxqs="8"
dev.ix.0.iflib.override_ntxqs="8"
dev.ix.1.iflib.override_nrxqs="8"
dev.ix.1.iflib.override_ntxqs="8"
dev.ix.0.iflib.separate_txrx="1"
dev.ix.1.iflib.separate_txrx="1"
dev.ix.0.iflib.rx_budget="65535"
dev.ix.1.iflib.rx_budget="65535"
dev.ix.0.iflib.override_nrxds="4096"
dev.ix.1.iflib.override_nrxds="4096"
dev.ix.0.iflib.override_ntxds="4096"
dev.ix.1.iflib.override_ntxds="4096"

inferno480

I appreciate all the support I received on here, but today I ended up returning my 8200. mostly it was me misunderstanding the advertised throughput values. I somehow either had the non-Firewalled or the TNSR values in my head which became my expectations. I was achieving 5Gb/s download speed pretty consistently, but needed a beefier box in order to exceed that so I'm going to be switching back to a DIY build (probably not as much overkill as I had before though).