[f-nsp] packetloss B2P622 POS blades

Jeroen Oldenhof jeroen at cj2.nl
Tue Feb 10 10:12:45 EST 2009


Ronald Esveld schreef:
> Jeroen,
>
> Try hw bc, to see if anything is wrong with the hw buffers
>   
Hi Ronald,

Thanks for your suggestion. I pasted the output of 'hw bc' and also 'sh 
back' for reference below.
I notice that the WriteDrops counters are rising pretty fast (1-10 
p/second) on the B24E at rtr-A and the B2P622 on rtr-B..

could this indicate a problem on these boards?

Best,
  - Jeroen


rtr-A#hw bc
Slot      FreeDepth  WriteDrop    WriteIn   WriteOut     ReadIn    ReadOut
                                    UseIn     UseOut     FreeIn    FreeOut
 1            3983    2942481   10814209   16571486   16571486   16571486
                                14751231    6051473    8993954    7662627
 2             900    8880735   11046211    5143640    5143640    5143640
                                 5143640    2165476   11046211   11046334
 4             910    3527799    4921759   11235329   11235329   11235329
                                11235329    1393960    4921759    4921872

rtr-A#hw back
Slot  Mod   FreeQ    DMADrop    BPDrop    WriteDrop     Last
-----------------------------------------------------------------------
 1   B4GMR4  3983       0          0    2942481     D:26 H:23M:24S:8
 2   B24E     900       0          0    8881143     D:26 H:23M:24S:8
 4   B2P622   910       0          0    3527799     D:26 H:23M:24S:8




rtr-B#hw bc
Slot      FreeDepth  WriteDrop    WriteIn   WriteOut     ReadIn    ReadOut
                                    UseIn     UseOut     FreeIn    FreeOut
 1            1939          7   10548638   14716992   14716992   14716992
                                10526350    6357989    6357996    4878671
 2             916          0    8861748    8861771    8861771    8861771
                                 8861772    8861749    8861749    8861856
 3             900          0     815540     815672     815672     815672
                                  815672     815540     815540     815663
 4             910     315291   15082124   16210906   16210906   16210906
                                16210906   14766833   15082124   15082237
rtr-B#sh back
Slot  Mod   FreeQ    DMADrop    BPDrop    WriteDrop     Last
-----------------------------------------------------------------------
 1   B8GMR   1939       0          0          7     D:1  H:15M:37S:5
 2   B8GC     916       0          0          0          NEVER
 3   B24E     900       0          0          0          NEVER
 4   B2P622   909       0          0     315580     D:1  H:15M:34S:5


> -----Oorspronkelijk bericht-----
> Van: foundry-nsp-bounces at puck.nether.net
> [mailto:foundry-nsp-bounces at puck.nether.net] Namens Jeroen Oldenhof
> Verzonden: dinsdag 10 februari 2009 14:29
> Aan: foundry-nsp at puck.nether.net
> Onderwerp: [f-nsp] packetloss B2P622 POS blades
>
> Hi!
>
> Has anyone experience with the B2P622 Foundry POS blades?
>
> We've deployed them around 9 months ago when we took our STM-1 in 
> production.. but since then we're constantly facing packetloss of about
> 1%.
>
> Our setup (hope it remains readable):
>                   +-------+    STM-1    +-------+
> Transits / <---+--+ RTR-A +=============+ RTR-B +--- <internal>
> IX's           |  +-------+             +-------+
>          +-----+-+
>          |monitor|
>          +-------+
>
> RTR-A&B: Foundry BI4000 w/MGMT4 ironcore running 07.8.04 (B2R07804).
> RTR-A is an edge-router facing several transits and exchanges. Through 
> the STM it is connected to RTR-B using iBGP.
> the monitor is used for smokeping and other management tasks.
>
> At first we performed eBGP routing to the ix's and transits on RTR-A. 
> This caused much packetloss, even when pinging from the monitor-box to 
> RTR-A directly. With no traffic packetloss is zero, but getting much 
> worse when the amount of traffic grows, around 4% at 60mbit. RTR-A was 
> also pulling around 20% CPU. I assume the POS interface has little or no
>
> CAM for layer 3, which makes it querying the CPU big time. This being a 
> resource consuming i guess some packets could be dropped there.
>
> We then moved all routing and BGP functionality to RTR-B, making RTR-A 
> simply a breakout box. Packetloss is reduced, but still around 1%.. and 
> the smokepings sill looks awful..
> We also swapped POS/FE/MGMT blades and ports and tried different 
> firmwares on both ends. No port-errors on both ethernet ports or POS 
> ports. The STM provider (TATA) reports no errors.
>
> On all paths outside the STM there is ZERO loss: from monitoring box to 
> several internet destinations, from and to several internal hosts past 
> the RTR.. so the STM and its interfaces are definately the problem.
>
> I figured out that some POS debugging can be done using 'dm console-on 4
>
> 2', 'dm cli 4 2 Q' returns some interesting commands.. but I can't find 
> a way to use them properly..
>
> so.. anyone on this list has any experience with these? Or encountered 
> simmillar issues?
>
> Thanks!
>
> Jeroen Oldenhof
>
> telnet at RTR-B# show pos 4/2
>
> POS4/2 is up, line protocol is up
>   No port name
>   Hardware is Packet over Sonet
>   Peer Internet address is 0.0.0.0
>   MTU 4470 bytes, encapsulation PPP, clock is line
>   Framing is SDH, BW 155000Kbit, CRC 32
>   Loopback not set, keepalive is set (10 sec), scramble disabled
>   LCP state is opened, IPCP state is init
>   300 second input rate: 50423416 bits/sec, 8213 packets/sec
>   300 second output rate: 12068536 bits/sec, 5802 packets/sec
>   2135136909 packets input, 11504550338524 bytes, 0 no buffer
>   Received 0 CRCs, 0 shorts, 0 giants, 0 alignments
>   1940413782 packets output, 2732631789028 bytes, 0 underruns
>   Line protocol is UP
>   Member of 5 L2 VLANs, port is tagged, port state is FORWARDING
>   STP configured to ON
>   Configured Path Trace String :
>   Received Path Trace String : RTR-A 4/2
>
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp
>   




More information about the foundry-nsp mailing list