[c-nsp] OSM memory allocating process question

[Rodney Dunn]

On Fri, Mar 31, 2006 at 01:26:27PM +0300, Emanuel Popa wrote:
> Hi,
> 
> We have a 7600 router positioned as border router. We have the full
> routing table there. One of the linecards in this router is a
> OSM-2OC48/1DPT-SS with 128MB of memory. Unfortunately for us, because
> of the large FIB:
> 
> CWTLC-Slot1#sh memory allocating-process totals | i FIB 1
> 0x604CD6A8   30718952     912  FIB 1 path chu
> 0x604A6E20     133880       4  FIB 1 path chunk pool

Yep that sounds about right..30M or so on the LC.

> 
> the OSM run out of memory and DCEF was disabled:
> 
> SLOT 4: Mar 31 07:19:34 UTC: %SYS-2-MALLOCFAIL: Memory allocation of
> 65536 bytes failed from 0x60170024, alignment 16
> SLOT 4: Mar 31 07:19:34 UTC: %FIB-3-NOMEM: Malloc Failure, disabling
> DCEF on linecard
> 
> ROUTER7600#sh cef linecard 4
> CEF linecard slot/cpu 4/0, status disabled
>  Sequence number 0, Maximum sequence number expected 0, Seq Epoch 3
>  Send failed 0, Out Of Sequence 0, drops 0, Wrong Epoch 0
>  Linecard CEF reset 2, reloaded 1
>  481596 elements packed in 38228 messages(19378197 bytes) sent
>  0 elements cleared
>  0/0/0 xdr elements in LowQ/MediumQ/HighQ
>  15293/53/503 peak elements on LowQ/MediumQ/HighQ
>  Keepalive timer is not running
>  Input  packets 25221870610, bytes 12653592553285
>  Output packets 2756416929, bytes 1804879415754, drops 0
> 
>  CEF Table statistics:
>  Table name                Version Prefix-xdr Status
>  Default                   11582484     474591 Active, table-disabled
> 
> But, the memory allocation on this linecard looks like this:
> 
> CWTLC-Slot4#sh memory allocating-process totals
>                 Head    Total(b)     Used(b)     Free(b)   Lowest(b)  Largest(b)
> Processor   6183D880    92022656    48681248    43341408      126888    11549520
>       I/O   20000000    16777216     3806952    12970264    12629072    12625400
>

Right..you got to 126k free to it ran out. Now that dCEF got disabled to it
you freed that 30+M back up so now free you have 43M (and the memory on the
OSM is pretty fragmented now because it got so low).

 
> Allocator PC Summary for: Processor
> 
>     PC          Total   Count  Name
> 0x603F4EC8   20447344       2  CWTLC STAT INFO
> 0x602D8740    6291568       2  SCP Find Master process
> 0x602D8A74    3604592       2  SCP Find Master process
> 0x603F4EB0    3145840       2  Init
> 0x602A0A20    1500056       1  rp if_info
> 0x604C18B8    1377432      21  Linecard CEF m
> 0x604597E4    1179648   12288  Init
> 0x603F2F2C    1080056       1  fr if_info
> [...]
> 
> Allocator PC Summary for: I/O
> 
>     PC          Total   Count  Name
> 0x600ECB70    2952320       5  COMM
> 0x600E6A54    1055392     338  *Packet Data*
> [...]
> 
> So, the machine allocates aprox. 87MB of memory for the processor and
> 16MB of memory for I/O opperations. I want to find out two things:
> 
> - is there any way to allocate more memory to the processor and less
> memory to the I/O?

I don't think that is possible.

> - can I see the rest of 25MB of memory and the allocation process?

The total is how much was left over to use after the image on the card
decompressed. S oyou has 92M left to run the services. 48M of that
was used the the processes before dCEF was on and took 40 more Meg.
You would have been running very close to the edge.

> 
> Of course, an upgrade is necessary and will be done eventually...

Yeah..but I do find it a bit odd that a STAT INFO type process would
be holding 20M. I think that is some information about QOS running
on the card. What code are you running?

I see some changes where they changed things in this area in 12.2(18)SXD
but I can't tell how much it would have changed it because it appears
to be configuration specific.

> Thank you,
> Emanuel Popa
> 
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/