tail drops on a vlan interface

[jigsaw2026]

I am getting large amounts of ingress packet drops on a vlan interface on a 3750. I affectively need to move the problem from there to the only physical interface in the vlan, so that I can apply L2 queueing (only thing available)...is there any way that I can increase the size of the buffer on the SVI? So that way the bottleneck occurs at the physical interface. Or anything else that can solve my problem?

Many thanks in advance,

J

[Mikhail.Galiulin]

Hi,

Check the nature of these drops. Larger buffer can help only in case traffic has short time peaks and rest of the time is on "normal" level. But if traffic is constantly on higher level than the switch can proceed than you need to increase capacity of the links.

//Mikhail Galiulin

[jigsaw2026]

Thanks Mikhail for your response.

The traffic is pretty much constantly higher, but here's what I don't understand; where would I need to increase capacity? On the link attached to the physical port? If so, why don't I see drops on the port itself? The vlan interface shows BW 1000000, whereas the physical is a 100MB link..so if anything, wouldn't the drops be on the egress queue of the vlan or the ingress of the physical?

Please help me, I have the feeling that I'm missing something fundamental here!

Many thanks, J

[Mikhail.Galiulin]

What kind of traffic are you talking about?

Is it traffic adressed to the switch?

//Mikhail Galiulin

[jigsaw2026]

Thank you Mikhail.

The traffic is regular IP traffic - mainly rdp, telnet, voice - which is destined for devices connected to the 3750 at other end of the link (the link being the only interface in the vlan in question). It's exactly the same set-up on that switch, whereby one physical port (the link) is in the vlan and then there's a vlan interface.

There would be very little traffic addressed to the switch.

[Mikhail.Galiulin]

So this VLAN is used only to connect two 3750 switches?

All your traffic (rdp, telnet, voice) is coming to the VLAN from other VLANs, right?

One more question:

Have you any empty ports on the switches?

//Mikhail Galiulin

[jigsaw2026]

That's correct (not my choosing).

Yep, roughly half are empty.

Thank you, much appreciated.

[Mikhail.Galiulin]

Try to configure Etherchannel between the switches...

http://www.cisco.com/en/US/tech/tk389/tk213/technologies_white_paper09186a0080092944.shtml

//Mikhail Galiulin

[jigsaw2026]

Thank you for your response.

I realise that increased bandwidth would relieve the congestion issue, but that is not an option for us.

What I would like to know is why the drops occur at the SVI level, and whether there's something I can do to cause the drops to occur at physical interface level (so that I can use LAN QoS to prioritise traffic)?

[Mikhail.Galiulin · Accepted solution]

Hi,

I'm not quite sure but I beleive that the drops occurs on SVI interface due to the following: when packet from one VLAN comes and the switch decides to route it to another VLAN it comes first internally to the SVI interface and then to the physical layer-2 interface (switchport). Packets comes to SVI interface faster than switchport can send them and congestion on the SVI interface occurs.

This is how I think the process looks like.

I don't think that you will make things much better if you implement QoS. QoS itself is of course very powerfull tool but it has its own limits. It can help in situation when you have a short time "peak" congestions. But if you have constant congestion the only ways to solve the problem are either increase the bandwidth or decrease the traffic.

Again, this is my vision of the problem. It could be that I don't know some important details which can lead to another recomendations.

//Mikhail Galiulin

[jigsaw2026]

Hi Mikhail,

You must be right in saying packets come to SVI interface faster than switchport can send them - I just thought that you would then see drops on the SVI's output or switchport's input queues.

Anyway, thank you very much for your thoughts.