In my experience with Nortel and Cisco switches we have so many
interface discards being reported that we tend to ignore interface
discards whenever and wherever they surface (by setting the "ignore
interface discards" behaviour). The discards happen in all situations
including under extremely light load.
I've opened cases with Cisco about this, and was told that the reason
discards were incrementing was a "cosmetic bug". That is, it didn't
reflect a lack of resources in the switches or dropped packets. Our
network performance tests indicate no packets are being dropped.
I think the discards may correspond to flooded packets where the
recipient switch believes that the egress port is the ingress port
(based on its forwarding table) and so drops the packet, although I
have no way of confirming this.
At 12:21 PM -0400 10/8/07, Richard E. Brown wrote:
--- Doug Veldhuisen wrote:
I am guessing since we are using Qos policies on these routers, I'm
seeing QoS drops, but not sure.
--- Christopher Sweeney wrote:
It's ifOutDiscards in the ifTable. In theory, I think, the discards
should indicate packets discarded because of things like resource
limitations. In practice, we have seen some SNMP implementations
which count other types of discards, such as discards due to firewall
rules, in this number.
--- end of quote ---
Yes, I have heard that some equipment may discard packets for "good
reasons" yet
count them in the MIB-II statistic ifOutDiscards. This could make
you think that
there's a problem, even though there isn't. Here are some examples:
- As Doug Veldhuisen notes above, a router may drop a packet
that's otherwise OK
because of a QoS policy. For example, if a stream of higher priority packets
needs to be transmitted, the router may drop the lower priority ones. These
packets could appear as discarded packets in MIB-II statistics.
- A firewall may also choose to count firewalled packets as
drops. This is an
explicit policy decision, not a lack of buffer space, but still
might be counted
as a MIB-II discard.
- Spanning Tree. Switches can (by design) ignore spanning tree
updates, because
they are not interesting or because spanning tree has been shut off. Certain
brands of switches can count these as MIB-II discards.
- Committed Information Rate. With permanent virtual circuits and similar
technologies, normal behavior of a customer device may be to attempt to send
data faster than the Committed Information Rate they have contracted for. The
router will drop those packets, and the drops may be counted in the
SNMP MIB-II
stats.
Any comments from the list on these or other situations? Thanks!
Rich Brown [EMAIL PROTECTED]
Dartware, LLC http://www.dartware.com
10 Buck Road, PO Box 130 Telephone: 603-643-9600
Hanover, NH 03755-0130 USA Fax: 603-643-2289
____________________________________________________________________
List archives:
http://www.mail-archive.com/intermapper-talk%40list.dartware.com/
To unsubscribe: send email to: [EMAIL PROTECTED]
____________________________________________________________________
List archives:
http://www.mail-archive.com/intermapper-talk%40list.dartware.com/
To unsubscribe: send email to: [EMAIL PROTECTED]
