packet loss on ME3400

[evan.r.moore]

I've been testing ME3400 switches at the bench, with results that concern me. I'm running very simple QoS, and when I put 45Mbps of UDP traffic through to a 100Mbps port, I get packet drops. The traffic is not very bursty, but I see drops whether the queue-limit is set low or high.

Has anyone else seen this sort of behavior? Is there a setting I've missed? IOS versions to avoid? Packet sizes which cause trouble (I test with 1400 byte packets to keep packet rates low)?

Any advice appreciated. I have a TAC case on this, but they have not been able to say much.

[sadbulali]

Use the troubleshooting guide present in the url given below for more info on troubleshooting packet loss issues in ME3400:

http://www.cisco.com/en/US/docs/switches/metro/me3400/software/release/12.2_44_se/configuration/guide/swtrbl.html

[DavidSarvai]

Hey Evan,

We are using the ME3400s now, and did run into some problems with QOS, as well as routing throughput on the smaller ME3400s running the enhanced Metro IOS.

We were in the lab using JPerf to generate traffic through the 3400s and found the smaller chassis maxed out below 40MB.

Drop me a line for more details.

I'm in NH.

[smebbin]

David,

I am having an issue where I am using Jperf to generate 100Mbps traffic on our links and the downstream comes back fine but the upstream is half or less of the downstream. The only thing I can think of is maybe asymmetrical routing? You have any ideas?

[ivan.gasparik]

Hi guys,

ME3400 has very small interface queues by default and therefore very sensitive to burstable traffic. Try to create a policy-map like:

policy-map fake

class class-default

queue-limit 544

and activate it on the interface:

interface FastEthernet0/1

service-policy output fake

Ivan

[Atif Siddiqui]

i dont see any interface drops, while having performance issues? should we implement this class-maps?

[Atif Siddiqui]

We are expereincing similar issues in our network; using ME switched to connec remote sites via TLS, having performance issues, asymmetric throughput using IPerf? on a 100 MB link we see 90 MB 1 directions and 55MB the other.

did you get any answer to this problem? we are using this code 12.2(35)SE1; cisco ME-3400G-12CS-D

[evan.r.moore]

I did not get any useful answers.

Sorry.

ERM

[Atif Siddiqui]

did you upgrade to fix? to what code.

thanks

[Atif Siddiqui]

I have got an answer today;

if you are going 1G to 100 MB on an ME 3400 swtich with an older code like 12.2(35)SE1; it will drop your performance to half. in other direction it won't as it is going to 100 MB to 1G...

new code 12.2(50)SE1 fixes it, i don't have the bug id as there are none, but will ask Cisco to file it.

[evan.r.moore]

Good job tracking that down, and thanks.

I've love to know the bug id when you get it.

ERM

[Atif Siddiqui]

Cisco has listed a cavet here:

Increased default value of buffer (from 48 to 160 packets - see Cisco

reference ID CSCsm80634)

which was the problem we were having. the queue size is increased in the new code.

[rkenny1976]

Hi Evan,

Can you share what the configuration is on the ME340x? I too, have ran into this issue in the past at the ISP. Thanks in advance for the clarification sir =))