ME3600X-24CX and high CPU usage due to Nile process

[DALIBOR TOMAN]

Hi,

if I insert SFP modules into ports CPU usage increases. Now I have 14SFP modules (mostly metalic ones) and the CPU is above 20% permanently. All of them used by Nile process. Which I understand is something which reads interface state and switches on/off the LED lights. I think that such a thing shouldn't eat 20+% of CPU power. There must be something wrong with the process.

As a result the switch is not able to respond to ICMP packets reliably (smokeping graph is very noisy). And I can imagine that other CPU tasks (BGP etc) could be affected too (but maybe the NILE process has lower priority).

Verified on another ME3600X-24CX and checked with various SFP modules.

Regards

Dalibor Toman

[Leo Laohoo]

And did you try to upgrade the IOS to something more recent?

[DALIBOR TOMAN]

sorry not mentioning it  - I upgraded to  15.3(3)S2 . The CPU usage didn't change.

I didn't try 15.3.(3)S3 yet ...

[Kanes Ramasamy]

Hi Dalibor,

Good day. 

Can i ask if you have managed to resolve the issue and if yes how? I am seeing a similar issue on my switch. 

Thanks in advance. 

Regards,

Kanes.R

[DALIBOR TOMAN]

I tried to solve the problem by opening a ticket with our vendor. But because Cisco doesn't treat the problem as a bug there is no real solution but a workaround only. We increased the NILE process poll period using this command:

test platform link_poll set_interval 80

The CPU usage lowered significantly. And the response time on smokeping graph is slightly better too (less higher response time peaks on the graph)

I have attached a graph of CPU usage for last 2 years. During may 2014 we replaced Cisco 3650G with the ME3600X and cpu usage went up. Next month we reconfigured the poll period and cpu went lower. During november we moved about 8 links to another device so the CPU usage lowered again.

Regards

D. Toman

[Kanes Ramasamy]

Hi Dalibor,

Thanks for the update and the fix. Really appreaciate that.

I will try to apply the fix on my router.

Regards,

Kanes.R

[golovchuk291]

I have the same problem on Cisco ME3800X. I have a problem solved by translation involved ports in the down mode in which were SFP modules. It is this process was taking 17% of the memory