em driver input errors

[alexpalias-bsdnet at yahoo.com]

--- On Thu, 8/20/09, Dmitriy Zamuraev <gigabyte.tmn at gmail.com> wrote:

> From: Dmitriy Zamuraev <gigabyte.tmn at gmail.com>
> Subject: Re: em driver input errors
> To: alexpalias-bsdnet at yahoo.com
> Cc: freebsd-net at freebsd.org
> Date: Thursday, August 20, 2009, 2:58 PM
> Hello Alex,
> 
> > SCHED_ULE, HZ=1000:
> I use this too
> 

> >  And "netstat -ni"
> shows 0 errors on all cards?
> Not exactly zero, but for uptime 155 days it seems to be
> ok.
> bras1 [/usr/home/dm]# netstat -i|grep em
> em0    1500 <Link#1>     00:15:17:71:f8:52 2457503820 20175 2096211799 0  0
> em1    1500 <Link#2>     00:15:17:71:f8:52 1084492221 11188 909418060    0 0
> em2    1500 <Link#3>     00:15:17:71:f8:52 4212941427 29566 3500442287 0    0
> em3    1500 <Link#4>     00:15:17:71:f8:52 2143321197     0 1878792786 0     0

Definitely way better than my numbers.

Name    Mtu Network       Address         Ipkts Ierrs    Opkts Oerrs  Coll
em0    1500 <Link#1>      blah1    11166666892 2776856 13194560132     0     0
em1    1500 <Link#2>      blah2    7420822703 462462 6185054971     0     0
em2    1500 <Link#3>      blah3    6070962640 167193 5341606634     0     0
em3    1500 <Link#4>      blah4    15183452     0 17424318     0     0

# uptime
 3:29PM  up 8 days, 11:19, 2 users, load averages: 0.69, 0.78, 0.83


> This counters was made by UDP flood, when dummynet can't
> process all packets and
> swi:net loads one core up to 100%. Yes, the dummynet on
> this machine, its bad idea but it's
> working stable. (After this incident the switches now
> control the flood attack)
> NOTE: the MAC is equal because i use if_lagg(4) for this
> interfaces for load all cores in CPU

Interesting.  But I'm seeing this during normal traffic.  I can't remember if it was this machine, or the older machine (also with em driver) when I did get a flood of over 90.000 packets/second, and the error rate jumped to around 50.000/second...

> >> I think it depends by motherbord, what full
> hardware specification are you using? with chips names
> > The machine is a Dell PowerEdge 2850.  According
> to its specs, the chipset is Intel E7520.
> > Two 64-bit Xeon processors at 3.20GHz, 4 GB RAM.
> For you bandwidth this server must work fine.
> Check the UDP/ICMP or other flood on em0 when errors
> appear.
> What kind of device at the end of em0 copper cable?
> If this a manageable switch, and supports tools - try to
> investigate what happens when errors appear.

It's a ZyXEL managed switch, unfortunately SNMP support doesn't seem to work very well on it.  But even before I added this switch (when there was a Cisco Catalyst 3560 at the end of the cable) I was seeing the error bursts.

But I will try and see if I detect anything on the switch end.

Thanks
Alex