ix hangs (almost)

[Gerrit Kühn]

Hi all,

Half a year ago we had reported here about issues with ix-interfaces when
transporting data over nfs. We only got about 50MByte/s back then, and
were not able to fix this issue. Instead, we went for rsyncd transport
that gave us about 200MByte/s (still not the full bandwidth, but
definitely better than nfs).

Now we recognize strange "hangs" in these sessions:
We do nightly backups on this link, and it works fine for days, sometimes
even weeks. Then suddenly the whole network traffic slows down almost to a
full halt. Write speeds (both rsyncd and nfs) go down to a few kByte/s. I
see no error messages, no bad packets, nothing that would indicate a reason
for this.
The issue goes away (for some time) when doing ifconfig ix0 down/up, but
comes back after some days.

Any suggestions on how to debug this (when it happens next) are highly
appreciated. If it looks like this is interface/driver-related, I'll also
take suggestions on what different 10GE (copper) card to try instead.


FreeBSD crest 10.2-RELEASE-p7 FreeBSD
10.2-RELEASE-p7 #0: Mon Nov  2 14:19:39 UTC 2015
root at amd64-builder.daemonology.net:/usr/obj/usr/src/sys/GENERIC  amd64


ix0 at pci0:5:0:0: class=0x020000 card=0x00028086 chip=0x15288086 rev=0x01
hdr=0x00    vendor     = 'Intel Corporation' device     = 'Ethernet
Controller 10-Gigabit X540-AT2' class      = network
    subclass   = ethernet


ix0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
options=8404bb<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,JUMBO_MTU,VLAN_HWCSUM,LRO,VLAN_HWTSO>
ether a0:36:9f:23:3d:b6 inet 192.168.6.2 netmask 0xffffff00 broadcast
192.168.6.255 nd6 options=29<PERFORMNUD,IFDISABLED,AUTO_LINKLOCAL> media:
Ethernet autoselect (10Gbase-T <full-duplex>) status: active



cu
  Gerrit