Where a buffer would be after being processed by the RP

[devinliu1991]

I've got a question about cisco buffer mechanism.

When a packet is waiting to be processed by the Routing Processor, it occupied specific buffer in the processor memory.

And my question is : after the packet is processed, will the buffer be added into the free list or just get released ?

[devinliu1991]

Added:

      will there be a process like that the IOS would see whether the free buffer count reaches the "max allowed" value if the newly free buffer is added into the free list ?

[Akash Agrawal]

Hi,

AFAIK there is interface buffer allocated to each interface for I/O traffic. When packets enters to the interface, it occupies the buffer and when leaves releases the buffer and same gets added to the free-list again otherwise free-list will keep on reducing and we will see buffer misses and packet drops. That could be condition during buffer leak. There is also process who checks on max-allowed value. Definition given below. You can also refer below link

https://netsuresolutions.zendesk.com/entries/20184378-Buffer-Leak-Symptoms-and-how-to-Identify-the-Nature-of-the-Error-

- MAX-ALLOWED: The maximum number of buffers in the free-list. If the number of

    buffers 'in free list' is greater than the 'max-allowed' value, the router will

    attempt to trim buffers from the pool. The 'max-allowed' parameter is used to

    prevent a pool from monopolizing buffers that it does not need anymore and free

    this memory back to the system for further use.

  - FREE-LIST: The number of buffers in the pool, ready for use.

- TRIMS: When the value 'in free list' exceeds that of 'max allowed' the processor

    trims the buffers.

HTH

Regards,

Akash

[devinliu1991]

Thanks for your reply.

When I had the checkup for the Cisco 2811, I got these output below:

show buffers 
Buffer elements: 
     1072 in free list (1119 max allowed) 
     104055923 hits, 0 misses, 619 created 
 
Public buffer pools: 
Small buffers, 104 bytes (total 191, permanent 50, peak 264 @ 7w0d): 
     21 in free list (20 min, 150 max allowed) 
     168577974 hits, 214695 misses, 36670 trims, 36811 created 
     4799 failures (0 no memory) 
Middle buffers, 600 bytes (total 25, permanent 25, peak 37 @ 7w0d): 
     15 in free list (10 min, 150 max allowed) 
     28090344 hits, 2522 misses, 259 trims, 259 created 
     1254 failures (0 no memory) 
Big buffers, 1536 bytes (total 50, permanent 50): 
     50 in free list (5 min, 150 max allowed) 
     4280928 hits, 0 misses, 0 trims, 0 created 
     0 failures (0 no memory) 
VeryBig buffers, 4520 bytes (total 10, permanent 10): 
     10 in free list (0 min, 100 max allowed) 
     0 hits, 0 misses, 0 trims, 0 created 
     0 failures (0 no memory) 
Large buffers, 5024 bytes (total 0, permanent 0): 
     0 in free list (0 min, 10 max allowed) 
     0 hits, 0 misses, 0 trims, 0 created 
     0 failures (0 no memory) 
Huge buffers, 18024 bytes (total 0, permanent 0): 
     0 in free list (0 min, 4 max allowed) 
     0 hits, 0 misses, 0 trims, 0 created 
     0 failures (0 no memory) 
 

The "Public buffer pools" refers to the process memory that handled by RP as I know.

Since most packets hit the Small buffers in the process memory, I think they belong to the control plane.

Now the device works normally both the hardware and system, and I could tell from the output that there are 170 buffers in Small buffers pool that are in use, 21 in the free list and consider these statistics are the normal state of the device.

But there was also a burst of traffic at 7w0d that pushed the RP to create 264 buffers and met a number of failures.

Is it right to say that there could probably be a burst of control plane traffic such as routing protocol convergence at 7w0d since the device started about 1 year ago?

And

If advices needed, though it might not be serious, is it right to give more Small buffers by tuning the min-free parameter to prevent "a might happen again burst" (the memory usage is about 7%)?