XConnect Performance Issues

[mmarchuk]

Has anyone got an idea on how much degradation I should expect on xconnect pseudowires. I have noticed a DRAMATIC difference in speed (being very careful to isolate it to just the xconnect part).

For example, on two ME6524's connected to each other with a full-duplex 100mb/s wire, and a server on each of the two ME6524's performing FTP on a standard layer 2 VLAN, I can get about 10-11MB/s throughput. Which is what I expect.

When I replace the VLAN's with an XCONNECT pseudowire tunnel, the performance drops to 200KB/s. I would expect SOME overhead, but this is terrible. I am currently in a lab, so I can make changes easily to test settings. If you want me to post the configs, I can. However, the configuration of the two interfaces is below:

Switch#1 (loopback IP 4.0.0.1)

interface GigabitEthernet1/17

no ip address

speed 100

duplex full

xconnect 4.0.0.5 17 encapsulation mpls

Switch#2 (loopback IP 4.0.0.5)

interface GigabitEthernet1/17

no ip address

speed 100

duplex full

xconnect 4.0.0.1 17 encapsulation mpls

Thanks in advance for your time!

[mheusing]

Hi,

One question: the MPLS MTU in the ME6524 is sufficient to transport the Ethernet frames?

Regards, Martin

[mmarchuk]

I have the MTU set to 1510 on the internal GigE ports (that are transport between the two switches)

[mheusing]

Hi,

if you transport ethernet frames with two labels (maybe one, because of PHP), then 1510 is not enough. Try 1600 and let us know, if the situation improved.

Regards, Martin

[mmarchuk]

Well, I tried 1600 and it did not improve the throughput. There does not appear to be excessive CPU utilization overhead for tunneling the packets, so I'm not sure where the slow down could be?

I am open to any suggestions.

[shivlu jain]

actuall this is very much true; becasue we have using the psuedowires in our network and there are overheads which pseudowires used to take. But if u r using encap mpls then its fine becasue in case of l2tpv3 you face lot of problems.

regards

shivlu

[mmarchuk]

Have you encountered any performance issues related to using pseudowires? What has your experience been using pseudowires?

[jon.harald.bovre]

Hi

You should verify CEF on router. Check CEF switching are running on MPLS interfaces. Bad performance could come from interface process switching.

Test with ping from workstations with DF set.

Verify MTU on router (system mtu, interface mtu)

Post 'show mpls l2 vc 17 detail' from both routers

I would not expect any performance drop when moving from VLAN to EoMPLS

Jon