outbound discards ???

[Stephane Caron]

We are getting "outbound discards" on our server and I am wondering what they are exactly and why we may be getting them.

Can anybody explain what a "outbound discard" is ? RFC1213's description of ifOutDiscards is:

"The number of outbound packets which were chosen to be discarded even though no errors had been detected to prevent their being transmitted. One possible reason for discarding such a packet could be to free up buffer space."

That's kind of vague... How can I pin-point my problem ?

Our server:

K580
lan card: Hewlett-Packard 10/100Base-TX Full-Duplex (ioscan type: btlan1)
connected to a 100baseT switch

output from lanadmin for this card:

LAN INTERFACE STATUS DISPLAY
Wed, Nov 7,2001 10:04:35

PPA Number = 2
Description = lan2 Hewlett-Packard 10/100Base-TX Full-Duplex
Manu Hw Rev 0
Type (value) = ethernet-csmacd(6)
MTU Size = 1500
Speed = 100000000
Station Address = 0x00108334919e
Administration Status (value) = up(1)
Operation Status (value) = up(1)
Last Change = 11548
Inbound Octets = 1275146304
Inbound Unicast Packets = 382248175
Inbound Non-Unicast Packets = 10248431
Inbound Discards = 204
Inbound Errors = 0
Inbound Unknown Protocols = 957999
Outbound Octets = 3478023079
Outbound Unicast Packets = 383630500
Outbound Non-Unicast Packets = 996546
Outbound Discards = 933593
Outbound Errors = 0
Outbound Queue Length = 0
Specific = 655367

Press to continue


Ethernet-like Statistics Group

Index = 2
Alignment Errors = 0
FCS Errors = 0
Single Collision Frames = 0
Multiple Collision Frames = 0
Deferred Transmissions = 0
Late Collisions = 0
Excessive Collisions = 0
Internal MAC Transmit Errors = 0
Carrier Sense Errors = 0
Frames Too Long = 0
Internal MAC Receive Errors = 0

Thanks for any help !

[Sanjay_6]

Hi Stephane,

Take a look at the thread below to know more about outbound discards"

http://us-support2.external.hp.com/cki/bin/doc.pl/sid=f73b1ba313c4c476db/screen=ckiDisplayDocument?docId=200000015662995

Hope this helps.

Regds

[Stephane Caron]

I did a search in ITRC and found no satisfactory information about "outbound discards" specifically... does anyone possess information on the specific "outbout discards" statistic ?

I see no reason why we should be getting some discarded packets outbound. This is a fully optimized switched environment so I doubt the network side is causing the problem.

One thing to note is that from what I can gather, these discards are occuring during our backups (we are gathing this info througout the day using Foglight, which is very handy...)

The feeling I have is that the Omniback Disk Agents are pushing out the info to the NIC faster than it can handle it, buffers get full, and the driver drops some of the outbound packets. But this would mean that NIC I/O is asynchronous.

But I can't confirm this, I am not an expert in NIC drivers. I am just looking for more explanation on what "outbound discards" really means. All explanations I can find are pretty vague therefore useless...

Thanks for any response !

[Anthony deRito]

An outbound discard is a packet that is not delivered. It does not mean that an error was detected but for some reason, possibly due to buffering or queuing issues, it was selected not to be transmitted to whatever higher layer protocol is was destined for. Then again, it really does not matter since TCP uses positive acknowledgement based on a 3-way handshake. TCP will retransmit the packet anyhow if it does not receive an expected acknowlegement. The problem is that if there are too many drops, TCP spends more time re-transmitting the packet than is productively worthwhile.

Tony

[Stephane Caron]

Your definition of outbound packet looks a lot more like the definition of inbound...

For me outbound packets are in transit to a lower layer, not a higher layer. And this makes a big difference. You are that an error is detected, possibly due to buffering or queuing issues. This is the specific info I am looking for. Relying on tcp re-transmit mechanisms to correct an underlying problem is more like a plaster, IMHO.

I guess I did not formulate my question properly. I really meant "is there any way to get more specific information on why the packets are being dropped". And also, "are there resources that need to be tuned, such as buffer sizes, etc, when I experience outbound packet drops."

I see no occurence of any outbound drops on any of our N-Class systems. I do understand they are faster machines. The NIC driver is totally different also, and may be handling situations differently.

In a nushell, I guess: there is a metric indicating a situation. There must be a reason for this metric to exist. What is it ?

[Sridhar Bhaskarla]

What kind of application does this box run?. If there are applications like firewalls etc., that do IP filtering, we can see the many outbound discards. Also, if the interface is down due to any reasons, you can see the discards. So,if the discards you are seeing are growing gradually it could be due to some other reason. If not, they might be during the time when the network was down.

-Sri

[Anthony deRito · Accepted solution]

With respect to the sending device, the higher layer refers to the higher layers of the destination device. A packet is encapsualted going down the stack on the sending device and decapsulated going up the stack on the receiving device. As it is decapsulated, it is destined for a higher layer protocol. When speaking about the TCP/IP stack, this is the TCP protocol at the transport layer of course. TCP is reliable transfer. It is the one to provide error detection as well as error correction... The lower layer protocols only provide error detection and leave it up to the higher layer protocols (TCP) to correct by resending.

As far as tuning, I am sorry I cannot help you.

Tony

[Stephane Caron]

I doubt that packets discarded by firewalls will show in the lan adapter RFC1213 stats... They will show in the firewall stats... As mentioned above, the discarded outbound packets stats seem to be showing up during backups, which is the time when the NIC card is most busy (it is handling around 11000 packets/sec, or 53 MBits/s, which is the most I have seen out of an HP-PB 100BT card... we do get over 90 MBits/s on ou N-Class systems, with no outbound drops...).

Otherwise, the system is handling a 3-tier ERP system with Tuxedo as the App Server engine and Oracle as the backend database. During the day, no packets get dropped. Only during backups. The drop rate is less than 1%, so I am not that concerned. I am just curious as to what I am supposed to do with that metric.

with regards to Anthony's response: I apologize. I did not mean to get into a definition debate. I am just trying to understand what is going on, and I tend to like details... I hope you were not offended by my response.