Catalyst 2960 1gbit uplink throughput problem

Unanswered Question
Apr 13th, 2010

Hi there

We have a wierd problem with our Catalyst 2960 switch.

The switch is connected to a 4510 switch at our provider via a 1000BASE-BS-D module. At the provider end and in the 2960 sh int says 1000 full duplex.

We havent done any settings on the port.

When we do a test internally on the switch we can create a throughput at 120/120 Megabyte a second which means full duplex and everything is fine.

When we go through our ISP we can have exactly 120 Megabyte a second in total. It looks like half duplex but link says full 1000.

Any suggestions

Regars

Carsten

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
jetnetshot Tue, 04/13/2010 - 12:09

Hi,

I might be missing something, if so correct me but it seems you are getting expected speeds?

2960 - local switch.

4510 - ISP switch.

Switches connected via fibre SFP - 1000BASE-BS-D.

Your tests give you 120MB/s (960Mbps ~ 1Gbps) speeds which is what is to be expected?

Does 'show int' show half duplex at either end?

How did you do your throughput tests?

One thing you might want to make sure, as good practice is to check and see the interface on each switch is configured appropriately for speed and duplex. It should either be auto speed and duplex on both sides or you should statically configure same speed and duplex on either end. You should not have one end manually configured for speed and duplex while the other end configured for auto speed and duplex since this will result in half duplex setting on the auto neg interface.

Regards,

Jet

Leo Laohoo Tue, 04/13/2010 - 15:17

Cisco SFP modules (except the GLC-T) will always negotiate to 1Gb.

lidldk000 Tue, 04/13/2010 - 22:19

Hi Expected is 120 MB/S up and 120 MB/S down at the same

time. We get 120 MB in total. Both ends can only autonegotiate and shows 1000 full doing sh int.

We have done a test internally on the 2960 between to linux server with RAM disk in order to secure that the error isnt locally.

When we do that we can get 120/120 instead of total 120

Regards

Carsten

jetnetshot Tue, 04/13/2010 - 23:26

Hi

If we have more details on how the test was done it may help. For example where was the traffic initiated from and where was it destined to. Is the end to end connection a tcp connection? I suppose you would have more control when testing locally rather than through your ISP. If traffic is flowing via the ISP switch are you sure that the path is at 1Gbps (that is from host A to host B).

I think the link is performing as it should at 1Gbps (according to 'show int', 1000 full). Lets assume you were doing TCP throughput tests between host A and host B locally. If you started connection from host A only OR host B only, you will see 120MB/s at each instance of the test. However if you simultaneously initiated download from both ends, .i.e host A and host B initiating connection to each other and downloading at same time, you will see that a single tcp connection from any of these host will only have rates of 60MB/s, half of 120MB/s, however the link will be fully utilised at 120MB/s with traffic flowing both ways simultaneously.

(If you are still suspicious of half duplex operation at either end, try the 'show controllers ..' command and see if you can spot any frame collisions, which is normal if any of the interface is operating at half duplex)

Regards,

Jet

Actions

This Discussion