Re: vge traffic problem

On Sun, Jan 10, 2010 at 11:04:08PM -0800, David Ehrmann wrote:
> Pyun YongHyeon wrote:
> >On Fri, Jan 08, 2010 at 02:43:02PM -0800, David Ehrmann wrote:
> >  
> >>It seems to be working, but I'm going to keep testing it before I 
> >>    
> >>proclaim it fixed.  Will these changes be in 8.1?
> >>    
> I spoke too soon.  Now an nfs client is having problems.  I ran iperf 
> and saw what looks like a decent rx connection, but a tx connection 
> where up to 2% of UDP packets are dropped (see 
> http://pastebin.com/m1c067417 ).  If both sides are gigabit connections, 
> I see more like 13%.  If I flip client and server sides, the results 
> still show issues with vge tx.
> 

Does netstat(1) also agree on the statistics?
Another thing you may want to see is hardware MAC statistics. See
the output of `sysctl dev.vge.0.stats` and check how many dropped
counters are changed.

> I ran iperf like this:  iperf -s -u -r and iperf -c 10.0.x.x -u -r.  
> Other machines have problems with the vge interface, but drop no more 
> than a packet between themselves.
> 
> This doesn't happen with my vr interface.  I gave it an IP in a 
> 192.168.x.x subnet and repeated the test.  Most of the time, zero 
> packets were dropped.  When packets were dropped, it was fewer than 2%.
> 

If receiver drops TX UDP frame sent by vge(4) would you try
disabling TX checksum offloading of vge(4)? If packet drop happens
only with UDP frames it could be checksum offload bug. Does your
controller is VT6130(PCIe)?

> Are there any known bugs that cause UDP tx packet drops on vge?

No, I'm not aware of this. If you can narrow down which packets
were dropped and can capture packets with tcpdump that would be
great help to analyze the issue.