On Tue, 26 Nov 2019 08:06:37 +0200 Artem Viklenko ar...@viklenko.net said

Sorry, small update.

Just re-cheked. It was not final change... wrong place. I've set it even smaller than 4096. Now it 3072.

Bummer. :(


Sorry.

No problem. Thanks for trying! :)

--Chris

26.11.19 07:55, Artem Viklenko via freebsd-net пише:
> Hi!
> > I have several small boxes with realtek nics acting as a router/firewall. > Also had same issues. FreeBSD driver didn't work at least for me so I > switched to Realtek's driver. But after some time traffic stops passing > my routers. Did some investigation and found that the issue is 9k mbufs.
> As far as I understand more traffic you push more issues with 9k mbufs
> appears due to memory fragmentation.
> You can check it with 'vmstat -z | grep mbuf'.
> > So I decided to do wery dirty hack - I've changed Jumbo_Frame_9k
> to Jumbo_Frame_4k in the if_re.c from Realtek's latest 1.95 driver.
> It comiles and work on FreeBSD 10.x and 11.x and now it just works
> (vendor says that driver is for older versions of the FreeBSD OS).
> And no more issues.
> > ITEM                   SIZE  LIMIT     USED    
> FREE      REQ FAIL SLEEP
> > mbuf_packet:            256, 2362080,       2,    1263,
> 2054916,   0,   0
> mbuf:                   256, 2362080,     514,   
> 1776,3460790080,   0,   0
> mbuf_cluster:          2048, 369076,    1265,      31, 
> 154081,   0,   0
> mbuf_jumbo_page:       4096, 184537,     513,    
> 294,1592339809,   0,   0
> mbuf_jumbo_9k:         9216,  54677,       0,      
> 0,       0,   0,   0
> mbuf_jumbo_16k:       16384,  30756,       0,      
> 0,       0,   0,   0
> > Now driver use mbuf_jumbo_page not mbuf_jumbo_9k and no fails. > > I'm ok with mtu 1500 in my environment and I don't know if mtu 9000 will
> work with this change. But at least it is stable now even after 100 days
> of uptime (just rebooted after upgraded to 11.3-RELEASE-p5).
> > Hope this helps. > > > 26.11.19 02:44, Chris пише:
>> Or at least make it non fatal.
>> OK here's the story; I'm experimenting with a multiport NIC (re(4))
>> as we hope to start using multiport 10G NICs.
>> Any of the re's we've used in the past have been very stable, which
>> is why I picked the one I did for this experiment. This one has been
>> performing rock solid for some 4 to 6 mos, under full time use. That
>> is until the last week. Where we're seeing:
>> watchdog timeout
>> repeated frequently. Which is ultimately fatal. ifconfig up/down will
>> not resuscitate it. Nor will service ifconfig restart, or plugging/
>> unplugging the cable(s). Bouncing the server is the only cure. Which
>> is unacceptable. Any, and All suggestions, or insight into the matter
>> GREATLY appreciated. Note; while this is an old 11.1, we're not planning
>> to up this box until we can confirm this can be cured. :)
>>
>> Details follow:
>> 11.1-STABLE r327867 amd64
>>
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>> watchdog timeout
>>
>> rc.conf(5)
>> ifconfig_re0="inet AA.BBB.CC.XX netmask 255.255.255.0 rxcsum txcsum tso4"
>> ifconfig_re1="inet AA.BBB.CC.WW netmask 255.255.255.0 rxcsum txcsum tso4"
>> ifconfig_re1_alias0="inet AA.BBB.CC.ZZ netmask 255.255.255.0"
>>
>> ifconfig(8)
>> re0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
>>
>      
options=8219b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM,TSO4,WOL_MAGIC,LINKSTATE>
> >>
>>      ether 00:13:3b:0f:13:44
>>      hwaddr 00:13:3b:0f:13:44
>>      inet6 fe80::213:3bff:fe0f:1344%re0 prefixlen 64 scopeid 0x1 >>      inet AA.BBB.CC.XX netmask 0xffffff00 broadcast 24.113.41.255 >>      nd6 options=23<PERFORMNUD,ACCEPT_RTADV,AUTO_LINKLOCAL>
>>      media: Ethernet autoselect (1000baseT <full-duplex>)
>>      status: active
>> re1: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
>>
>      
options=8219b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM,TSO4,WOL_MAGIC,LINKSTATE>
> >>
>>      ether 00:13:3b:0f:13:45
>>      hwaddr 00:13:3b:0f:13:45
>>      inet AA.BBB.CC.WW netmask 0xffffff00 broadcast 24.113.41.255 >>      inet AA.BBB.CC.ZZ netmask 0xffffff00 broadcast 24.113.41.255 >>      inet6 fe80::213:3bff:fe0f:1345%re1 prefixlen 64 scopeid 0x2 >>     nd6 options=23<PERFORMNUD,ACCEPT_RTADV,AUTO_LINKLOCAL>
>>      media: Ethernet autoselect (1000baseT <full-duplex>)
>>      status: active
>> lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> metric 0 mtu 16384
>>      options=600003<RXCSUM,TXCSUM,RXCSUM_IPV6,TXCSUM_IPV6>
>>      inet6 ::1 prefixlen 128     inet6 fe80::1%lo0 prefixlen 64 >> scopeid 0x3     inet 127.0.0.1 netmask 0xff000000     nd6 >> options=21<PERFORMNUD,AUTO_LINKLOCAL>
>>      groups: lo
>> pciconf(8)
>> re0@pci0:5:0:0:    class=0x020000 card=0x012310ec chip=0x816810ec >> rev=0x07 hdr=0x00
>>     vendor     = 'Realtek Semiconductor Co., Ltd.'
>>     device     = 'RTL8111/8168/8411 PCI Express Gigabit Ethernet >> Controller'
>>     class      = network
>>     subclass   = ethernet
>> re1@pci0:6:0:0:    class=0x020000 card=0x012310ec chip=0x816810ec >> rev=0x07 hdr=0x00
>>     vendor     = 'Realtek Semiconductor Co., Ltd.'
>>     device     = 'RTL8111/8168/8411 PCI Express Gigabit Ethernet >> Controller'
>>     class      = network
>>     subclass   = ethernet
>>
>> Thanks again!
>>
>> --Chris
>>
>>
>> _______________________________________________
>> freebsd-net@freebsd.org mailing list
>> https://lists.freebsd.org/mailman/listinfo/freebsd-net
>> To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
>
--
Regards!
_______________________________________________
freebsd-net@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"


_______________________________________________
freebsd-net@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-net
To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"

Reply via email to