are dropped packets a sign of network problem?

[progressdll]

in a xen environment , i see a lot op dropped packets via netstat -i

Is this a sign of network problems, or is it normal to see this kind of
numbers? i'm not sure how to interprete the data. is this normal, bad,
critical. What are your stats on this?

I guess i have a xen issue of some sort, but not sure what yet

# netstat -i
Kernel Interface table
Iface       MTU Met    RX-OK RX-ERR RX-DRP RX-OVR    TX-OK TX-ERR TX-DRP
TX-OVR Flg
eth0       1500   0 53999491      0      0      0  6234874      0      0     
0 BMRU
lo        16436   0  9247734      0      0      0  9247734      0      0     
0 LRU
peth0      1500   0 135679270      0      0      0 95455766      0      0     
0 BORU
tap0       1500   0  5411876      0      0      0 35071803      0      0     
0 BMRU
vif-vmdeb  1500   0        0      0      0      0       35      0 333350     
0 BORU
vif0.0     1500   0  6234874      0      0      0 53999492      0      0     
0 BORU
vif6.0     1500   0 55813641      0      0      0 98482968      0 482338     
0 BORU
vif6.1     1500   0        0      0      0      0       39      0 46875771     
0 BORU
vif68.0    1500   0        0      0      0      0        0      0      0     
0 BOU
vif69.0    1500   0  9349348      0      0      0 30571290      0   3282     
0 BORU
vif69.1    1500   0        0      0      0      0       44      0 18947590     
0 BORU
vif70.0    1500   0   129605      0      0      0   460256      0    702     
0 BORU
vif70.1    1500   0        0      0      0      0       35      0 333350     
0 BORU
xenbr0     1500   0 46298695      0      0      0        0      0      0     
0 BORU

[Gabriel M. Schuyler]

We've had some network problems that were resolved by disabling tcp  
checksum offloading on all the domUs.  Our symptom was we couldn't  
ssh (or do much else) between two domUs on the same physical host.

So on each domU we run:

sudo /sbin/ethtool -K eth0 tx off

Hope this helps!
Gabe

[Daniel P. Berrange]

On Fri, Oct 12, 2007 at 10:35:25AM -0700, Gabriel M. Schuyler wrote:
> We've had some network problems that were resolved by disabling tcp  
> checksum offloading on all the domUs.  Our symptom was we couldn't  
> ssh (or do much else) between two domUs on the same physical host.

That should be neccessary at all - there were some bugs in this respect
a long time ago, but it works fine now, assuming you keep your kernels
up2date with errata. Turning off checksum offload will destroy network
performance/throughput for your guests - particularly guest<->guest
networking, but also guest -> network.

Dan.