cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Dropped packets on AMD

gregg_kniss
Guide

We are seeing approximately 2k/hr of "dropped packets" and equally "packets not received by driver". I checked several other "error" metrics for the interfaces and they are all 0. I worked with the our network team to check the switch ports and they are not reporting any incrementing errors or overruns and the switch CPU is relatively low. The utilization is low at 2-3Mbps on a 1Gbps link. Below is the definition from the documentation as to these metrics.

Dropped packets

The number of packets dropped by the interface
while reading packets from a physical connection (wire).

Packets not received by driver

The sum of interface received overruns.

Is there a better definition of these to metrics? Based on these defnintions I'm not really sure where to look. The AMD and switch appear to be fine.

7 REPLIES 7

ulf_thornander3
Inactive

What version?

What NIC and driver?

Is the AMD "Clean" - nothing else running like services or AV?

12.2.3

Intel PRO/1000 and e1000e_rtm

Yes, nothing else running.

chris_v
Dynatrace Pro
Dynatrace Pro

That's terrible, I've seen even the most under-resourced virtualised AMD easily cope with much much more than 2-3Mbps.

raffaele_talari
Inactive
@Gregg Kniss

could you provide additional info about it?

Erik_Soderquist
Dynatrace Pro
Dynatrace Pro

Sounds like some form of bottleneck strangling to me.

How many decodes are in use on this AMD? Balanced usage or mostly focused on one or two decodes? How much RAM and how is the Shared Memory Buffer configured?

-- Erik





















6.042% Generic (with transactions)
0.002% HTTP
14.796% SOAP over HTTPS
0.001% SSL Decrypted
79.159% TDS

RAM is 128GB and I've gone through rtminst and configured everything to its recommended value.

ulf_thornander3
Inactive

So it's in SQL somewhere. I would try to get a trace of the traffic to the SQL servers and just look at the traffic if something sticks out.