[c-nsp] GSR E2 linecard performance woes

Oliver Boehmer (oboehmer) oboehmer at cisco.com
Wed Mar 16 07:58:35 EST 2005



> The following "bad" event counters are observed to increment largely:
> 
> - "Packets punt to RP" (does this mean RP CPU or LC CPU?)

Generally the RP, this is stuff going to the router itself.. Usually the
RP takes care of these control-plane packets, but sometimes the LC
engine does (like answering "pings").

> - some "HW engine reject" (not much)

Ok, so this doesn't point to a problem.

> Where can I find a list of which features appear as PSA bundles for
> this card in different IOS releases?
> (trying to avoid parsing *all* the release notes)

Hmm, I don't think there is a document describing this in necessary
details.. Those bundles don't usually change that much, it could be that
we introduce new bundles as new features come along..

> For instance, I note that "output_acl_128" is not enabled,
> do I need this to maintain performance with egress ACLs?

Ah.. Do you have Engine 3 or 4+ LC in your chassis? Then it could be
that the LC-CPU is busy processing the egress ACL. The PSA is not
available in the egress path. Take a look at
http://www.cisco.com/warp/public/63/acl_12000.html for details on how
ACLs are processed on the GSR..

Otherwise it would be better to contact TAC since we might need more
detailed info to troubleshoot what's going on exactly..

	oli

> 
> Oliver Boehmer (oboehmer) wrote:
>> David,
>> 
>> E2 generally forwards in hardware, so LC-CPU usage is no straight
>> indication for LC utilization. LC-CPU is invoked when the hardware
>> punts traffic to the LC-CPU, so you want to find out what is
>> happening here. Does your loaded PSA bundle support all the features
>> you have configured on your LC (the PSA ASIC doing the forwarding is
>> loaded with a microcode bundle depending on the feature you have
>> enabled. If the loaded bundle does not support all the features,
>> pkts are punted to the LC-CPU for switching).. 
>> 
>> "exec slot 5 sh contr psa bund"
>> "exec slot 5 sh contr events" (several times to see which counter
>> increases) 
>> 
>> might provide more info..
>> 
>> 	oli
>> 
>> 
>> David Freedman <> wrote on Wednesday, March 16, 2005 12:32 PM:
>> 
>>> Hiya,
>>> 
>>> AFAIK, the cisco "marketing" packet rate figure for the GSR120XX E2
>>> linecards is 4MPPS. 
>>> 
>>> I'm running a 3GE E2 linecard in a 12012 with the following
>>> features: 
>>> 
>>> 
>>> Global -
>>> 
>>> access-list compiled
>>> 
>>> 
>>> Port 1 -
>>> 
>>> - Ingress ACL (7 lines)
>>> - Egress ACL (7 lines)
>>> - netflow sampling (interval = 1/1000)
>>> - sparse pim
>>> - 120MB/Sec ingress
>>> - 190MB/Sec egress
>>> - 39Kpps ingress
>>> - 34Kpps egress
>>> 
>>> 
>>> Port 2 -
>>> 
>>> - Ingress ACL (7 lines)
>>> - Egress ACL (7 lines)
>>> - netflow sampling (interval = 1/1000)
>>> - sparse pim
>>> - 320MB/Sec ingress
>>> - 320MB/Sec egress
>>> - 49Kpps ingress
>>> - 77Kpps egress
>>> 
>>> Port 3 -
>>> 
>>> - Ingress ACL (7 lines)
>>> - Egress ACL (7 lines)
>>> - netflow sampling (interval = 1/1000)
>>> - sparse pim
>>> - MPLS (tag-switching) with NO TE
>>> - 532MB/Sec ingress
>>> - 513MB/Sec egress
>>> - 120Kpps ingress
>>> - 97Kpps egress
>>> 
>>> 
>>> The same ACL is used both ingress and egress on all ports.
>>> 
>>> 
>>> Now look at this:
>>> 
>>> #execute-on slot 5 sh proc cpu | exc 0.00
>>> ========= Line Card (Slot 5) =========
>>> 
>>> CPU utilization for five seconds: 94%/73%; one minute: 85%; five
>>>   minutes: 84% PID Runtime(ms)   Invoked      uSecs   5Sec   1Min  
>>>     5Min TTY Process 4   172497132   1297743     132922  1.53% 
>>>    0.51%  0.55% 0 BFLC PSA ACL 10    59944212  62202985        963 
>>> 0.07%  0.14% 
>>> 0.13%   0 CEF LC
>>> IPC Backg
>>>    21   114470916  13418571       8530  0.23%  0.23%  0.22%   0
>>>    Per-Second Jobs 46    43622128  66352970        657  0.23% 
>>>    0.29%  0.29%   0 Queue Mgr 59  2177980408   6086199     357858
>>> 18.19%  7.89%  7.70%   0 TAG Stats Backgr
>>> 
>>> 
>>> Why is the card working so hard?
>>> Is the marketing data that far off?
>>> or am I doing something known to degrade the performance of the
>>> card? 
>>> 
>>> Removing the netflow and PIM doesn't really affect the CPU
>>> interrupt percentage much..... 
>>> 
>>> Here is the sh diag for the slot:
>>> 
>>> 
>>> 
>>> SLOT 5  (RP/LC 5 ): 3 Port Gigabit Ethernet
>>>    MAIN: type 68,  800-6376-05 rev B0
>>>          Deviation: 0
>>>          HW config: 0x00    SW key: 00-00-00
>>>    PCA:  73-4775-07 rev C0 ver 2
>>>          Design Release 2.0  S/N XXXXXXXXXX
>>>    MBUS: Embedded Agent
>>>          Test hist: 0x00    RMA#: 00-00-00    RMA hist: 0x00
>>>    DIAG: Test count: 0x00000003    Test results: 0x00000000
>>>    FRU:  Linecard/Module: 3GE-GBIC-SC=
>>>          Route Memory: MEM-GRP/LC-256=
>>>          Packet Memory: MEM-LC1-PKT-256=
>>>    L3 Engine: 2 - Backbone OC48 (2.5 Gbps)
>>>    MBUS Agent Software version 1.86 (RAM) (ROM version is 2.32)   
>>>    ROM Monitor version 16.12 Fabric Downloader version used 9.3
>>>    (ROM version is 9.3)    Primary clock is CSC 0 Board is analyzed
>>>    Board State is Line Card Enabled (IOS  RUN )
>>>    Insertion time: 00:09:58 (21w6d ago)
>>>    DRAM size: 268435456 bytes
>>>    FrFab SDRAM size: 134217728 bytes, SDRAM pagesize: 8192 bytes
>>>    ToFab SDRAM size: 134217728 bytes, SDRAM pagesize: 8192 bytes   
>>> 0 crashes since restart 
>>> 
>>> 
>>> 
>>> System is running  12.0(25)S4 and all linecards had firmware
>>> upgraded via the "upgrade all" command.
>>> 
>>> Processor is a fully expanded GRP-B.
>>> 
>>> Any help (or even confirmation that I am running the card hot)
>>> would be appreciated. 
>>> 
>>> 
>>> Thanks,
>>> 
>>> Dave.
>>> 
>>> 
>>> _______________________________________________
>>> 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/
>> 
>> _______________________________________________
>> 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/
>> 
> 
> _______________________________________________
> 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/



More information about the cisco-nsp mailing list