Problem with NICs flapping at intervals of 5 mins
-
@whosmatt Can you even set manual 10ge, thought part of the spec was auto? You can set the speed down manual.. I have to call up the spec.. And we really don't run much copper 10ge at work.. I believe we do have some.. I will have to tool around tmrw and see..
Might be able to set it 5 or 2.5, etc.
-
@johnpoz said in Problem with NICs flapping at intervals of 5 mins:
Can you even set manual 10ge
It's an option in the UI, yes. It's also an option on the switch side.
@johnpoz said in Problem with NICs flapping at intervals of 5 mins:
And we really don't run much copper 10ge at work
I'm beginning to understand why.
-
@johnpoz said in Problem with NICs flapping at intervals of 5 mins:
Might be able to set it 5 or 2.5, etc.
I tried setting 5000Base-T and ifconfig still shows "media: Ethernet autoselect (10Gbase-T <full-duplex>)"
If I go through the various settings on the switch the NIC follows along, down as far as 1Gbps. There's also a 100M Full setting on the switch but the NIC won't link at that speed.
-
The options offered in the gui are what ifconfig -m returns. For example:
[admin@7100.stevew.lan]/root: ifconfig -vvm ixl0 ixl0: flags=1008843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST,LOWER_UP> metric 0 mtu 1500 options=48100b8<VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,VLAN_HWFILTER,HWSTATS,MEXTPG> capabilities=4f507bb<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,TSO4,TSO6,LRO,VLAN_HWFILTER,VLAN_HWTSO,NETMAP,RXCSUM_IPV6,TXCSUM_IPV6,HWSTATS,MEXTPG> ether 00:e0:ed:86:a6:8c inet6 fe80::208:a2ff:fe0e:a591%ixl0 prefixlen 64 scopeid 0x1 media: Ethernet autoselect (10GBase-AOC <full-duplex>) status: active supported media: media autoselect nd6 options=21<PERFORMNUD,AUTO_LINKLOCAL> drivername: ixl0 plugged: SFP/SFP+/SFP28 1X Copper Active (Copper pigtail) vendor: BROCADE PN: 58-1000026-01 SN: CAX116410001093 DATE: 2016-10-07
DACs like that usually don't offer more than one speed.
-
@stephenw10 so yeah that doesn't show any options.. My igb0 on the other hand does
media: Ethernet autoselect (1000baseT <full-duplex>) status: active supported media: media autoselect media 1000baseT media 1000baseT mediaopt full-duplex media 100baseTX mediaopt full-duplex media 100baseTX media 10baseT/UTP mediaopt full-duplex media 10baseT/UTP
-
Yeah I can see the list of supported speed / duplex, it's just that setting any of them in the UI doesn't seem to change the media from autoselect:
ixl1: flags=1008843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST,LOWER_UP> metric 0 mtu 1500 description: OPT10 options=48100b8<VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,VLAN_HWFILTER,HWSTATS,MEXTPG> capabilities=4f507bb<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,TSO4,TSO6,LRO,VLAN_HWFILTER,VLAN_HWTSO,NETMAP,RXCSUM_IPV6,TXCSUM_IPV6,HWSTATS,MEXTPG> ether b4:96:91:b6:27:b5 inet6 fe80::b696:91ff:feb6:27b5%ixl1 prefixlen 64 scopeid 0x2 media: Ethernet autoselect (1000baseT <full-duplex>) status: active supported media: media autoselect media 10Gbase-T media 5000Base-T media 2500Base-T media 1000baseT media 100baseTX nd6 options=21<PERFORMNUD,AUTO_LINKLOCAL>
I've got it running at 1000Mbps right now because the port is forced to that speed on the switch. Incidentally, it hasn't gone down since I set that about 12 hours ago or so.
-
Hmm, I wonder it's misreporting 'autoselect' there. If the switch side is set to 1G fixed the NIC should not be able to negotiate with it.
I guess we'll see if it makes any difference anyway.
-
@stephenw10 said in Problem with NICs flapping at intervals of 5 mins:
Hmm, I wonder it's misreporting 'autoselect' there. If the switch side is set to 1G fixed the NIC should not be able to negotiate with it.
I guess we'll see if it makes any difference anyway.
I'm wondering if the setting on the switch side is really forcing speed/duplex or just forcing it to auto negotiate to a predetermined speed. If that makes sense. In other words it's still autoselect, but the list of possible values has been narrowed.
-
Yes, that could certainly be the case.
You can set that in pfSense using sysctl:
[admin@7100.stevew.lan]/root: sysctl -d dev.ixl.0.advertise_speed dev.ixl.0.advertise_speed: Control advertised link speed. Flags: 0x1 - advertise 100M 0x2 - advertise 1G 0x4 - advertise 10G 0x8 - advertise 20G 0x10 - advertise 25G 0x20 - advertise 40G 0x40 - advertise 2.5G 0x80 - advertise 5G Set to 0 to disable link. Use "sysctl -x" to view flags properly.
But if it is still negotiating and doesn't lose link at 1G that might also be a clue.
-
@stephenw10 said in Problem with NICs flapping at intervals of 5 mins:
Hmm, I wonder it's misreporting 'autoselect' there. If the switch side is set to 1G fixed the NIC should not be able to negotiate with it.
I guess we'll see if it makes any difference anyway.
When in doubt, look at the logs I guess. I should have looked sooner:
2024-10-15 15:29:09.000 kernel: kernel: ixl1: Media change is not supported. 2024-10-15 15:29:09.000 php-fpm[28896]: php-fpm[28896]: /interfaces.php: The command '/sbin/ifconfig 'ixl1' media '10Gbase-T'' returned exit code '1', the output was 'ifconfig: SIOCSIFMEDIA (media): Operation not supported by device'
-
@whosmatt said in Problem with NICs flapping at intervals of 5 mins:
but the list of possible values has been narrowed.
Yeah that very well could be - since gig came out, the preferred setting has been auto neg.. If you want a slower speed you should really just limit what is offered, etc.
While forcing with gig is an option, the best practice is this is only for temp troubleshooting to figure out why auto isn't working.. Been a really long time I have looked at the actual spec for 10ge, but I recall reading that there would be no option for hard set, had to be auto..
Makers don't always follow the spec ;) heheh but if your running fine on gig - but having issues with 10ge over copper - how long are the runs? Any way you could switch to fiber?
I would have to do some looking around on work network.. Only been at this gig going on a year and some of these locations were inherited when a company was bought, etc. We still have some old HPs that working on getting rid of, etc. But I believe the only copper 10ge we have is from the server in the rack to the TOR switch in that same rack. And the only reason those are copper is the server came with the 10ge copper connections or something.. We sure do not run any switch to switch over copper that I am aware of. But there are a lot of sites ;) and I haven't even been on some of the switches or routers in those sites..
-
@johnpoz said in Problem with NICs flapping at intervals of 5 mins:
Any way you could switch to fiber?
I intend to switch to DAC. I was using the X710-T4L because my AT&T fiber CPE has a multigig port (up to 5Gbps) and I could never get any igc card stable. I have since bypassed the AT&T CPE by using a SFP+ module that has the ONT built in to it. Right now that SFP+ module is plugged into a dumb switch acting as a media converter but my intention is to get a X520-DA2 and just go direct fiber for WAN and DAC for all the inside interfaces.
-
Ok, so I plugged ixl1 into a different switch at 10Gbps and the problem followed. Gonna force it down to 5Gbps and see if it does any better there.
-
So far so good at 5Gbps. To answer the earlier question about cabling, pfSense is connected directly to the switch with 14' Monoprice Slimrun Cat6A patch cables (orange if that matters).
-
Still good at 5Gbps. What I did, for anyone curious, is force my speeds at the switch, since they don't appear to apply on the pfSense side. Not complaining about pfSense since it can only support what the hardware supports. After evaluating ixl1 at 5Gbps on a different switch, I reconfigured my lagg on the pfSense side to two NICs (ixl0 and ixl1) and on the switch to port 7 and port 8. On the switch side, my Trendnet TEG-7124WS allowed me to to set the port speed on the "trunk" (their nomenclature for link aggregation) and it forced both ports down to 5Gbps and the ports on pfSense followed suit. So I'm currently at a 2 port lagg with each port auto negotiating to 5Gbps.
I'm starting to think that 10Gbps over twisted pair copper is a crapshoot, and it's not just because of this experience. I had an event last night with my TrueNAS box where both NICs in a lagg experienced up/down events about 10 minutes apart and I logged (smokeping) some packet loss. The up/down events on that box are rare and i've never seen any packet loss so going to keep watching that.
-
One more update relevant to this subject:
I checked a couple of pfSense VMs we have at work. Those have passed through X550 NICs and have also had negotiation issues. Setting the X550 to 10Gbase-T does in fact work there on the pfSense side so I have to assume it's something specific to the X710 or to the ixl driver.
-
Mmm, does seem like it's the driver. I expected that to work. I don't have anything base-T to test it with here currently I know it did work with X500 based NICs. Especially the X550-T that can do n-base speeds.
-
@stephenw10 said in Problem with NICs flapping at intervals of 5 mins:
Especially the X550-T that can do n-base speeds.
Yeah, that's what we have. The switches, on the other hand, are older Cisco Nexus 5000 series that don't support the intermediate speeds. We were seeing speed mismatches that caused all sorts of problems as you might imagine.
-
One more update on this subject. pfSense is still stable with zero up/down events since limiting to 5Gbps.
My TrueNAS box, on the other hand, started showing the same behavior (at the same 5 min intervals) after a software update. They were happening about 1x per day, but even though only one interface in the lagg would go down at any given time, I'd measure some packet loss, likely because smokeping is also on a 5 min interval. This was a change from previous observed behavior where the up/down events would occasionally happen, but not on the 5 min interval and no measured packet loss.
I picked up a Dell X710-T2L for a song on ebay (seriously, like $30 shipped) and cross-flashed it to the Intel firmware. I replaced the X540-T2. My intention was to also limit to 5Gbps but I haven't had to do that. The events just stopped and 0% loss since then. It's even more puzzling because the cabling here, while short, has multiple excursions into wall-mounted jacks with Cat6 (not Cat6a) running under the house through the crawl space. That was not the case with pfSense, which is directly cabled to the switch with Cat6a patch cables.
Don't really know what to think about this other than that 10GBase-T is a crapshoot.