[c-nsp] Outbound drops on 6748

[Matthew Huff]

Different architecture. It's designed to handle bursty traffic. The output buffer overflow on any hardware can be caused by the output buffer filling up when a microburst overflows the queue faster than it can serialize the output on the wire. The end host speed is irrelevant. It can't get it out on the copper/fiber fast enough. The solutions are 1) Increase the speed of the output (1 -> 10GB, 10GB->40GB) 2) Use QOS to shape the traffic 3) Use bigger port buffers (in some cases a smaller shared buffer > bigger individual buffers)



 But I would definitely talk to your VAR/rep. Also look at Arista.


> -----Original Message-----
> From: Robert Hass [mailto:robhass at gmail.com]
> Sent: Saturday, January 28, 2012 12:52 PM
> To: Matthew Huff
> Cc: cisco-nsp at puck.nether.net
> Subject: Re: [c-nsp] Outbound drops on 6748
> 
> On Sat, Jan 28, 2012 at 6:42 PM, Matthew Huff <mhuff at ox.com> wrote:
> > Cisco Nexus 3000 Series switches. They came out to compete with Arista in the HFT world,
> but are useful anywhere latency and/or bursting is an issue:
> >
> > http://www.cisco.com/en/US/products/ps11541/index.html
> 
> Nexus 3000 have 9MB buffers comparing to 1.3MB per port at
> WS-X6748-GE-TX. Will ultra-low-latency switching decreases need amount
> of buffers ? What about for using Nexus 3000 for long-distance
> connection (eg. 120KM). Do I need more buffers for 120KM Ethernet that
> for 10KM Ethernet ? (normal traffic , no storage)