[c-nsp] 7500 queue question

Rik Koenig rkoenig at rivulet.com
Mon Jun 20 19:12:17 EDT 2005


I was playing around with the tx-XXX-limit commands, and the tx-ring-limit
didn't seem to do a whole lot, but I must admit my understanding of all the
mechanisms in place leave something to be desired. Is there a diagram of the
transmit/receive path going through a router available?

The tx-queue-limit command however, had a dramatic impact on what we saw.
Running a PQ with a limit of 20, and a tx-queue-limit of single digits (I
could only get it down to 4, even though the command indicates you can go
down to 1) and we're seeing packet drops at bursts of 24. Smaller bursts,
like 23, seem to also drop packets, but much more rarely.

Yeah, I suppose expecting a 1:1 correlation is unrealistic; it was just a
bit shocking to see up to 170 fit in a 40.

One other question I have is, what is the default for the tx-queue-limit? I
looked around through documentation and google, but couldn't find what the
default level is.

Thanks,

~Rik

-----Original Message-----
From: Rodney Dunn [mailto:rodunn at cisco.com] 
Sent: Monday, June 20, 2005 6:37 PM
To: Rik Koenig
Cc: 'Rodney Dunn'; cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] 7500 queue question

I'm not sure but see if you can turn the tx-ring-limit down
and see if that gives you closer correlation to your output
queue depth for drops.

It does sound like the output queue depth is independent
of the tx ring depth but I could very well be wrong there.

Expecting a 1:1 correlation for a burst and drops is
a pretty difficulat thing to test I would think because you could
process some packets while cycling through to process the rx/tx
rings of other interfaces.



On Mon, Jun 20, 2005 at 01:11:14PM -0400, Rik Koenig wrote:
> So, tx ring is 128 deep, if I am reading this correctly?
> 
> 
> Interface Ethernet1/1
> Hardware is AMD Unknown
> ADDR: 60BCD700, FASTSEND: 6015AC04, MCI_INDEX: 29
> DIST ROUTE ENABLED: 0
> Route Cache Flag: 1
>  LADRF=0x0000 0x0100 0x0000 0x0000
>  CSR0  =0x00000072, CSR3  =0x00001044, CSR4  =0x0000491D, CSR15
=0x00000000
>  CSR80 =0x0000C800, CSR114=0x00000000, CRDA  =0x30049FC0, CXDA
=0x3004A170
>  BCR9 =0x00000001 (full-duplex)
>  HW filtering information:
>   Promiscuous Mode Disabled, PHY Addr Enabled, Broadcast Addr Enabled
>   PHY Addr=00E0.14BA.C729, Multicast Filter=0x0000 0x0100 0x0000 0x0000
>  amdp2_instance=0x60BCF420, registers=0x52804000, ib=0x30049E60
>  rx ring entries=32, tx ring entries=128
>  rxring=0x30049EC0, rxr shadow=0x60BCF680, rx_head=16, rx_tail=0
>  txring=0x3004A100, txr shadow=0x60BCF740, tx_head=7, tx_tail=7,
tx_count=0
> 
>  VIP addr_filter=0x60BCFB80, af_index=0, af_entries=3, af_stat=0x0007
>   Addr=00E0.14BA.C729, Mask=0000.0000.0000, Filter=0x1100
>   Addr=FFFF.FFFF.FFFF, Mask=0000.0000.0000, Filter=0x0200
>   Addr=0100.0CCC.CCCC, Mask=0000.0000.0000, Filter=0x0200
>  pak_to_host=0x602ADD40, vip_memd_ifcntl=0x60952CD0
>  spurious_idon=0, filtered_pak=0, throttled=0, enabled=0, disabled=0
>  rx_framing_err=0, rx_overflow_err=0, rx_buffer_err=0, rx_bpe_err=0
>  rx_soft_overflow_err=0, rx_no_enp=0, rx_discard=0, rx_miss_count=0
>  tx_one_col_err=0, tx_more_col_err=0, tx_no_enp=0, tx_deferred_err=0
>  tx_underrun_err=0, tx_late_collision_err=0, tx_loss_carrier_err=3781
>  tx_exc_collision_err=0, tx_buff_err=0, fatal_tx_err=0
>  hsrp_conf=0
>  tx_limited=0(128)
> 
> -----Original Message-----
> From: Rodney Dunn [mailto:rodunn at cisco.com] 
> Sent: Monday, June 20, 2005 12:28 PM
> To: Rik Koenig
> Cc: 'Rodney Dunn'; cisco-nsp at puck.nether.net
> Subject: Re: [c-nsp] 7500 queue question
> 
> if-con <slot>
> sh controller
> 
> On Mon, Jun 20, 2005 at 11:41:31AM -0400, Rik Koenig wrote:
> > I get no information on the e/fe interfaces using 'show contr'. I'm
using 
> > 12.2(21a).
> > 
> > Inter packet gap is 960ns. 
> > 
> > 
> > -----Original Message-----
> > From: Rodney Dunn [mailto:rodunn at cisco.com] 
> > Sent: Monday, June 20, 2005 10:52 AM
> > To: Rik Koenig
> > Cc: cisco-nsp at puck.nether.net
> > Subject: Re: [c-nsp] 7500 queue question
> > 
> > What is the depth of the controller txring in 'sh contr'?
> > 
> > I can't remember if the output hold queue is inclusive of packets that
> > have already been placed on the txring of the interface to go out.
> > I would think it is conclusive.
> > 
> > You have to understand your traffig generator and how it generates
> > burst also. ie: what is the inter-packet gap.
> > 
> > Another thing that is complicated is the servicing
> > of rx and tx rings and how a few packets will sit on the rx rings until
> > the interrupt service routine gets a chance to do the lookup to move
> > it to the tx side.
> > 
> >  
> > 
> > 
> > On Mon, Jun 20, 2005 at 09:56:28AM -0400, Rik Koenig wrote:
> > > Hello all,
> > > 
> > >  
> > > 
> > > Last week we were playing around with output queue depths and packet
> loss.
> > > Running a 7507 with RSP4, and VIP2-50 (128/8), and PA-FE-TX and PA-4E.
> > Using
> > > the 100Mb Ethernet as input and the 10Mb as output. CEF is running,
but
> > not
> > > dCEF. 
> > > 
> > >  
> > > 
> > > We would burst a given number of 224B frames into the 100Mb port, to
> exit
> > > the 10Mb port. Given the 40-deep FIFO exit queue, we needed to burst
170
> > > packets before any were dropped! When we dropped the queue down to 5
or
> 1,
> > > we needed 40 or so. 
> > > 
> > >  
> > > 
> > > Now, to my thinking, if a 100Mb link is receiving 41 packets, a 10Mb
> link
> > > that reports it can hold 40 packets should back up and tail-drop. But
> this
> > > didn't happen. What am I missing?
> > > 
> > >  
> > > 
> > > Thanks,
> > > 
> > > ~Rik
> > > 
> > 
> > > _______________________________________________
> > > 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