Packet Errors and Drop

Unanswered Question

I'm looking at the Cisco Network Assistant Tool, for errors and drop packets.

I noticed the tool show per individual switch an amount of 20,000 packets drops and in some other switches up to 35,000 packets drops.

I checked both G0/1 and router interface (fiber optic connection) for speed difference.

Both sides are setup for 1000 and full duplex.

Is there a reason for so many packet drops, and/or solution to minimize this issue?


  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 5 (1 ratings)
Reza Sharifi Sat, 11/07/2009 - 20:18
User Badges:
  • Super Bronze, 10000 points or more
  • Cisco Designated VIP,

    2017 LAN


Do you any QOS, Policing, shaping configured on these interface?


Joseph W. Doherty Sun, 11/08/2009 - 04:34
User Badges:
  • Super Bronze, 10000 points or more

A count of packets drops, alone, tells us little. What's we need to know is drops vs. total traffic number of packets. A very rough rule of thumb is only become concerned with a drop rate that exceeds 1%.

Consider if you had a 24 FastEthernet port switch with a gig uplink, more than 10 ports transmitting concurrently will send more data than the gig uplink can forward. Initially switch will queue excess traffic but eventually queue will fill and then result is drops.

Correction and/or minimization of such drops depends on the nature of the congestion. If in the prior exampe, congestion was very transient, increasing buffer or queue allocation might preclude drops although at the risk of additional queuing latency. If the congestion is sustained, more bandwidth is necessary to avoid the congestion.

QoS techniques can be used to either minimize percentage of drops for TCP traffic (e.g. RED and/or tail drop per flow) or preclude drops on some traffic while other traffic perhaps has even more drops (e.g. VoIP vs. FTP).

(BTW, often egress queuing is single interface FIFO which is often far from optimal for minimizing drops.)

Joseph W. Doherty Sun, 11/08/2009 - 05:24
User Badges:
  • Super Bronze, 10000 points or more

If my message seemed to convey that you're overreacting, such was not intended. Your second post about wanting to "improve network efficiency" is one I use myself.

Again, you'll first need to ascertain whether this number of drops is really an issue. There's a big difference between 20 K drops from 200 K packets vs 2,000 M packets.

Even when there's not a real cause for concern, and then "improve network efficiency" might come into play, you have to consider cost vs. benefit. If the situation is 20 K drops over 2,000 M packets, and you can drop (pun not intended) drops to 19 K by activation of one QoS command or by a very, very complex QoS configuration, the former might be worthwhile the latter might not.

What I would suggest, is first determine % of drops. If critical, try to mitigate ([edit] NB: on LANs, increased bandwidth capacity is often an attractive solution). If not critical, optionally, determine whether they can be mitigated and consider cost vs. benefit in doing so.


This Discussion