[c-nsp] Curious... 2800 or 3800

[Justin Shore]

That's 520Mbps of 64 byte packets on a NPE-G1 *doing absolutely nothing 
else.*  No ACLs, NAT, PPPoE, EGP, IGP, NetFlow, NBAR, SNMP, uRPF, CoPP, 
QoS, CAR, logging of any kind, heaven forbid any crypto, etc, yadda, 
yadda.  Not even someone sitting on the console running sh proc cpu to 
monitor the load (pull the 5min avg instead).  Everything done in HW and 
virtually nothing switched across the CPU.  Nothing a router under a 
normal workload in a normal configuration would be doing.  That's how 
it's been explained to me.  I've even seen an SE jot down some figures 
on a napkin that basically showed how ASIC XYZ has a bus ABC wide.  At 
DEF Mhz it can process a header in GHI amount of time which means it can 
process JKL headers per second.  Assuming a minimum size payload of MNO 
the ASIC can switch amount of data at a rate of PQR per second. 
Assuming that the CEF entry had already been made and that there wasn't 
going to be any incurred overhead from processing switching the first 
packet.  And assuming no other delays in the switching process.  His 
number did roughly jive with the router performance doc.

I would be interested to have better clarification from Cisco on the 
test environment but I would say that it is highly unlikely that 
high-overhead transport mediums and ATM and POS were used for the test. 
  It would be interesting to see a parallel version of 
routerperformance.pdf with a basic configuration that involves a lot of 
the "normal" stuff I mentioned above.  I'd be very interested in seeing 
the comparison.  Where else but in a lab environment will we ever see 
the best case scenario?  I don't think I've ever built such a scenario 
myself. :-)

Justin


jim bartus wrote:
> Thats 520mbps of 64 byte packets on the NPE-G1.  "Normal" workloads 
> (like say IMIX) could easily see 4 - 5 times that.
> 
> -jim
> 
> On 4/2/07, *Justin Shore * <justin at justinshore.com 
> <mailto:justin at justinshore.com>> wrote:
> 
>     Gary Stanley wrote:
>      > At 04:29 AM 4/2/2007, Shaun wrote:
>      >> The 2800's look to support a bunch of modules, ds1,ds3,gb,etc...
>     I was
>      >> wondering how well a 2811 or higher unit would work to use as a
>     border
>      >> router uplinked to a upstream at Gbit.  If the 2800's cant
>     handle it,
>      >> how about the 3800's?  Right now i'm using 3750's for this using
>     roughly
>      >> ~100mbit in/out and only receiving default routes from upstream.
>      >>
>      >> ~Shaun
>      >
>      > A 2811 will blow up with that much traffic for routing. I don't think
>      > any of the 28xx series can handle that much traffic, I had a 2851 and
>      > it was at 90% CPU with 75mbit of traffic
>      >
>      > Please read;
>      >
>      >
>     http://www.cisco.com/warp/public/765/tools/quickreference/routerperformance.pdf
> 
>     Agreed.  Pick up a 7206VXR w/ a NPE-G2 or a short stack 7600 if you want
>     to support that kind of bandwidth.  Even the 3845 maxes out at around
>     256MB.  A NPE-G1 is 520 I believe.  That NPE-G2 is a full Gig.
> 
>     Justin
>     _______________________________________________
>     cisco-nsp mailing list  cisco-nsp at puck.nether.net
>     <mailto:cisco-nsp at puck.nether.net>
>     https://puck.nether.net/mailman/listinfo/cisco-nsp
>     archive at http://puck.nether.net/pipermail/cisco-nsp/
> 
> 
> ------------------------------------------------------------------------
> NOTE: This message was trained as non-spam. If this is wrong, please 
> correct the training as soon as possible.
> Forget previous vote