Try lowering your mtu to 1500, that worked miracles for me..

-- 
Mark Schouten
Tuxis Internet Engineering
m...@tuxis.nl / 0318 200208

> On 25 May 2015, at 09:36, "Cs" <bim...@field.hu> wrote:
> 
> Hi all,
> 
> I have two FreeBSd 10.1-RELEASE servers connected to each other. They were 
> connected via cross link, but they are connected to a cisco switch now (the 
> problem was the same with cross link too). When transferring huge files 
> (50-500GB backup files) via Gigabit (it is important!) the network randomly 
> dies. The backup runs every day/week and sometimes the connection is ok for 
> months sometimes it happens twice a week. When the network dies I can log in 
> to the server via IPMI and use the console everything is OK, but can't send 
> anything out on the network. ifconfig em0 down/up doesn't help nor netif 
> restart. The problem never occured when I used 100Mbit connection between 
> them, but it was 3com NIC (xl), gigabit adapter is Intel (em0). When I limit 
> the transfer rate (rsync bandwith limit or ipfw pipe) the problem is much 
> more rare.
> 
> I tried to set these tuning parameters on both servers with different buffer 
> size but nothing helped:
> 
> # cat /etc/sysctl.conf
> security.bsd.see_other_uids=0
> net.inet.tcp.recvspace=512000
> net.route.netisr_maxqlen=2048
> kern.ipc.nmbclusters=1310720
> net.inet.tcp.sendbuf_max=16777216
> net.inet.tcp.recvbuf_max=16777216
> kern.ipc.soacceptqueue=32768
> 
> # cat /boot/loader.conf
> geom_mirror_load="YES" # RAID1 disk driver (see gmirror(8))
> ipfw_load="YES"
> net.inet.ip.fw.default_to_accept=1
> kern.maxusers=4096
> accf_data_load="YES"
> 
> The duplex settings are identical on both servers.
> 
> Server A:
> em1: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 9000
> options=4219b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM,TSO4,WOL_MAGIC,VLAN_HWTSO>
>  
>        ether 00:25:90:24:52:66
>        inet x.x.x.x netmask 0xfffffe00 broadcast x.x.x.x
>        nd6 options=29<PERFORMNUD,IFDISABLED,AUTO_LINKLOCAL>
>        media: Ethernet autoselect (1000baseT <full-duplex>)
>        status: active
> 
> Server B:
> em0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 9000
> options=4219b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM,TSO4,WOL_MAGIC,VLAN_HWTSO>
>  
>        ether 00:30:48:dd:fe:3e
>        inet x.x.x.x netmask 0xfffffe00 broadcast x.x.x.x
>        nd6 options=29<PERFORMNUD,IFDISABLED,AUTO_LINKLOCAL>
>        media: Ethernet autoselect (1000baseT <full-duplex>)
>        status: active
> 
> Today I tried to set mtu to 9000 but in tcpdump I see that during scp it is 
> still 1500:
>    x.x.x.x.222 > x.x.x.x.37612: Flags [.], cksum 0xb6ee (incorrect -> 
> 0xda6f), seq 35749, ack 113701596, win 7986, options [nop,nop,TS val 
> 3103966325 ecr 853712893], length 0
> 09:27:33.912354 IP (tos 0x8, ttl 64, id 1028, offset 0, flags [DF], proto TCP 
> (6), length 1500)
> 09:27:33.912358 IP (tos 0x8, ttl 64, id 1029, offset 0, flags [DF], proto TCP 
> (6), length 1500)
> 
> 
> Any ideas? Thanks guys!
> _______________________________________________
> freebsd-net@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-net
> To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"

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

Reply via email to