em driver input errors

[Artis Caune]

2009/8/1  <alexpalias-bsdnet at yahoo.com>:
> Good day
>
> I'm running a FreeBSD 7.2 router and I am seeing a lot of input errors on one of the em interfaces (em0), coupled with (at approximately the same times) much fewer errors on em1 and em2.  Monitoring is done with SNMP from another machine, and the CPU load as reported via SNMP is mostly below 30%, with a couple of spikes up to 35%.
>
> Software description:
>
> - FreeBSD 7.2-RELEASE-p2, amd64
> - bsnmpd with modules: hostres and (from ports) snmp_ucd
> - quagga 0.99.12 (running only zebra and bgpd)
> - netgraph (ng_ether and ng_netflow)
>
> Hardware description:
>
> - Dell machine, dual Xeon 3.20 GHz, 4 GB RAM
> - 2 x built-in gigabit interfaces (em0, em1)
> - 1 x dual-port gigabit interface, PCI-X (em2, em3) [see pciconf near the end]
>
>
> The machine receives the global routing table ("netstat -nr | wc -l" gives 289115 currently).
>
> All of the em interfaces are just configured "up", with various vlan interfaces on them.  Note that I use "kpps" to mean "thousands of packets per second", sorry if that's the wrong shorthand.
>
> - em0 sees a traffic of 10...22 kpps in, and 15...35 kpps out.  In bits, it's 30...120Mbits/s in, and 100...210Mbits/s out.  Vlans configured are vlan100 and vlan200, and most of the traffic is on vlan100 (vlan200 sees 4kpps in / 0.5kpps out maximum, with the average at about one third of this).  em0 is the external interface, and its traffic corresponds to the sum of traffic through em1 and em2
>
> - em1 has 5 vlans, and sees about 22kpps in / 11kpps out (maximum)
>
> - em2 has a single VLAN, and sees about 4...13kpps both in and out (almost equal in/out during most of the day)
>
> - em3 is a backup interface, with 2 VLANS, and is the only one which has seen no errors.
>
> Only the vlans on em0 are analyzed by ng_netflow, and the errors I'm seeing have started appearing days before netgraph was even loaded in the kernel.
>
> Tuning done:
>
> /boot/loader.conf:
> hw.em.rxd=4096
> hw.em.txd=4096
>
> Witout the above we were seeing way more errors, now they are reduced, but still come in bursts of over 1000 errors on em0.
>
> /etc/sysctl.conf:
> net.inet.ip.fastforwarding=1
> dev.em.0.rx_processing_limit=300
> dev.em.1.rx_processing_limit=300
> dev.em.2.rx_processing_limit=300
> dev.em.3.rx_processing_limit=300
>
> Still seeing errros, after some searching the mailing lists we also added:
>
> # the four lines below are repeated for em1, em2, em3
> dev.em.0.rx_int_delay=0
> dev.em.0.rx_abs_int_delay=0
> dev.em.0.tx_int_delay=0
> dev.em.0.tx_abs_int_delay=0
>
> Still getting errors, so I also added:
>
> net.inet.ip.intr_queue_maxlen=4096
> net.route.netisr_maxqlen=1024
>
> and
>
> kern.ipc.nmbclusters=655360
>
>
> Also tried with rx_processing_limit set to -1 on all em interfaces, still getting errors.
>
> Looking at the shape of the error and packet graphs, there seems to be a correlation between the number of packets per second on em0 and the height of the error "spikes" on the error graph.  These spikes are spread throughout the day, with spaces (zones with no errors) of various lengths (10 minutes ... 2 hours spaces within the last 24 hours), but sometimes there are errors even in the lowest kpps times of the day.
>
> em0 and em1 error times are correlated, with all errors on the graph for em0 having a smaller corresponding error spike on em1 at the same time, and sometimes an error spike on em2.
>
> The old router was seeing about the same traffic, and had em0, em1, re0 and re1 network cards, and was only seeing errors on the em cards.  It was running 7.2-PRERELEASE/i386
>
>
> Any suggestions would be greatly appreciated.  Please note that this is a live router, and I can't reboot it (unless absolutely necessary).  Tuning that can be applied without rebooting will be tried first.


Is it still actual?
You didn't mention if you are using pf or other firewall.
I have similar problem with two boxes replicating zfs pools, when I
noticed input errors.
After some investigation turns out it was pf overhead, even though I
was skipping on interfaces where zfs sedn/recv.

With pf enables (and skip) I can copy 50-80MB/s with 50-80Kpps and
0-100+ input drops per second.
With pf disabled I can copy constantly with 102 or 93 MB/s and
110-131Kpps, few drops (because 1 CPU almost eaten).





-- 
Artis Caune

    Everything should be made as simple as possible, but not simpler.