I have a MTR running , now my MPLS for the network starts at hop 5, strangely the latency gone up quite higher

MPLS enabled: latency on hop 5 is 23.5ms
HOST: Loss% Snt Last Avg Best Wrst StDev

2. 10.249.5.198 0.0% 2 0.5 0.3 0.2 0.5 0.2
3. 10.249.2.180 0.0% 2 0.9 1.1 0.9 1.2 0.2
4. 10.249.4.25 0.0% 2 1.8 1.8 1.8 1.8 0.0
5. 10.249.5.236 0.0% 2 23.5 23.8 23.5 24.2 0.5
6. 10.249.4.196 0.0% 2 27.0 28.8 27.0 30.5 2.4
7. 10.249.2.57 0.0% 2 24.0 25.5 24.0 26.9 2.0
8. 10.249.5.108 0.0% 2 25.5 24.2 22.9 25.5 1.9
9. 10.249.5.116 0.0% 2 33.6 28.1 22.5 33.6 7.8
10. 10.249.4.108 0.0% 2 28.2 25.1 21.9 28.2 4.4
11. 10.249.4.116 0.0% 2 22.4 24.1 22.4 25.9 2.5
12. 10.249.0.78 0.0% 2 21.3 19.6 17.9 21.3 2.4

example: MPLS LDP disabled on hop 5 and 6, latency down to 6.1ms

2. 10.249.5.198 0.0% 2 0.2 0.2 0.2 0.2 0.0
3. 10.249.2.180 0.0% 2 0.9 0.9 0.9 1.0 0.1
4. 10.249.4.25 0.0% 2 2.1 2.8 2.1 3.5 1.0
5. 10.249.5.236 0.0% 2 6.1 4.6 3.1 6.1 2.2
6. 10.249.4.196 0.0% 2 10.0 9.5 9.0 10.0 0.7
7. 10.249.2.57 0.0% 2 12.3 10.4 8.5 12.3 2.7
8. 10.249.5.108 0.0% 2 40.3 45.3 40.3 50.2 7.0
9. 10.249.5.116 0.0% 2 27.9 26.5 25.2 27.9 1.9
10. 10.249.4.108 0.0% 2 33.0 30.5 28.0 33.0 3.5
11. 10.249.4.116 0.0% 2 33.5 31.3 29.1 33.5 3.1
12. 10.249.0.78 0.0% 2 25.1 20.8 16.5 25.1 6.1

Notice the difference in latency, almost tripple less when disabled, what can cause this? MTU? i have no issues, just curious why it goes up , thought mpls improves switching?

the backbone kit is combination of trango, ubnt, radwin gear all with highest values mtu set in the radios, most RouterOS is combination of RB1200 / 493g / RB1100

tracepath with mpls enabled

2: 10.249.5.198 (10.249.5.198) 0.431ms
3: 10.249.2.180 (10.249.2.180) 1.400ms
4: 10.249.4.25 (10.249.4.25) 2.514ms
5: 10.249.5.236 (10.249.5.236) asymm 17 23.223ms
6: 10.249.4.196 (10.249.4.196) asymm 16 27.763ms
7: 10.249.2.57 (10.249.2.57) asymm 15 29.635ms
8: 10.249.5.108 (10.249.5.108) asymm 14 26.390ms
9: 10.249.5.116 (10.249.5.116) asymm 13 25.330ms
10: 10.249.4.108 (10.249.4.108) asymm 12 25.691ms
11: 10.249.4.116 (10.249.4.116) 26.117ms
12: 10.249.0.78 (10.249.0.78) 39.472ms reached
Resume: pmtu 1500 hops 12 back 12



Thanks

Sew

Hi..

could you please to reduce the the MPLS-MTU from 1508 to 1500...

/mpls interfaces set mpls-mtu=1500

let me know the result

Ali.

I lowered as suggested , here is the tracepath results, pmtu now 1496, this is mind boggling for me, I suspect this is causing issues on the voip under load , any other suggestions, the mpls forwarding table is about 1900 entries

hop 4 is a rb1100 with latest version and default mtu's

wireless is ubnt gear with mtu set to 2024

hop 5 is a rb493g with latest version and default mtu's

wireless is radwin 2000b which supports jumbro frames

hop 6 is a rb1200 ...

2: 10.249.5.198 (10.249.5.198) 0.443ms
3: 10.249.2.180 (10.249.2.180) 1.487ms
4: 10.249.4.25 (10.249.4.25) 2.588ms
5: 10.249.5.236 (10.249.5.236) asymm 17 27.560ms
6: 10.249.4.196 (10.249.4.196) asymm 16 19.516ms
7: 10.249.2.57 (10.249.2.57) asymm 15 24.922ms
8: 10.249.5.108 (10.249.5.108) asymm 14 26.663ms
9: 10.249.5.116 (10.249.5.116) asymm 13 29.884ms
10: 10.249.4.108 (10.249.4.108) asymm 12 26.084ms
11: 10.249.4.116 (10.249.4.116) 38.773ms
12: 10.249.0.78 (10.249.0.78) 27.650ms reached
Resume: pmtu 1496 hops 12 back 12

Hi..

I case ubnt make some latency in the frame delivery.. some thing in framming and deframming..

change from small to biggest mtu..!!!

as you see there latency is increased only between hop 4 to 5 that is ubnt in between..!!!

could you do some test from hope 5 to upper router..!! and c ..

this way its will be more clear where is the bottle neck ..!!!


Ali.

I had to put to rest an rb450g when I started using mpls / vpls because of mtu problems. Sounds like you could have mtu issues with your 493. I can't remember the specifics but seems all the 4xx series rb's could pass large enough frames to run mpls.

Okay i will have to dig a bit deeper after hours again...

If i go to each hop and ping the peers, the latency is 1ms, but only on the forwarding its causing the latency. i suspect the 4xx version is doing this, also for mikrotik enigineers, the ports is both on port 8 on the 1100 and the 493g, I know there was a bug fix in the 1200 series port 8 which solved my previous mtu issue, could be related to the 1100 or 493?

S

On MPLS enabled network packets are switched along the label switching path not directly back to the sender so it will increase the latency in pings and traceroutes.

Here is the ilustrations from cisco article:


http://www.cisco.com/en/US/tech/tk436/t ... a42a.shtml

It affects also ICMP error messages, you can read about it in the manual
http://wiki.mikrotik.com/wiki/Manual:MP ... LS_network

Okay even though latency is higher, normal packet forwarding should still be faster?, so you say icmp will only be affected?

S

MPLS forwarding will be faster than routing. You can't use ICMP to reliably determine link quality based on latency.

You can reliably use ICMP echo's but not any traceroute feature reliably.
I suspect the UBNT gear is causing your problems but it's very hard to say.
We have not been able to reliably use UBNT gear for MPLS transports.

We have not been able to reliably use UBNT gear for MPLS transports.

So I cannot use a UBNT Wireless transparent Bridge for MPLS labeld traffic?
(Like Nanobridge M5)