Back pressure from a SPAN port

[STUART KENDRICK]

Apparently, on the N5K, the SPAN function can induce back-pressure on the ingress interface.

Does this happen with any other platform?  Searching www.cisco.com, seems to me that this only applies to the N5K ... but perhaps this effect has only been /documented/ for the N5K ... perhaps it happens on other platforms.  Does anyone know?

Extracted from:

http://www.cisco.com/en/US/customer/docs/switches/datacenter/nexus5000/sw/system_management/521_n1_1/b_5k_System_Mgmt_Config_521N11_chapter_010000.html

"If the SPAN source interface sends more than  6-Gbps traffic or if traffic bursts too much, the device drops traffic  on the source interface. You can use the switchport monitor rate-limit 1G command on the SPAN destination to reduce the dropping of actual  traffic on the source interface; however, SPAN traffic is restricted to 1  Gbps."

"SPAN traffic is rate-limited as follows on Nexus 5500 series switches to prevent a negative impact to production traffic:

• SPAN is rate-limited to 5 Gbps for every 8 ports (one ASIC).
• RX-SPAN is rate-limited to 0.71 Gbps per port when the RX-traffic on the port exceeds 5 Gbps."

--sk

Stuart Kendrick

FHCRC

[STUART KENDRICK]

I've been poking at this question intermittently ... and I'm stumbling in finding a definitive source 'knowledge base' article.

Still, at the moment, it seems to me that the N5K has a unique feature here, the ability to shield ingress (source) traffic from the contention induced by a SPAN port.

I don't see a similar feature on the C4K or C6K, for example (i.e. rate-limiting SPAN ports).

In thinking about this, it seems to me that one needs to understand the architecture of the switch in order to assess the potential for impacting source traffic.  So, for example, if one is SPANning a port on a typical C4K line card (8 10/100/1000 ports sharing a single 1Gps uplink to the Sup card) *and* one is mirroring to another member of the same 8 port group, then it seems like to me that the source port would discard frames, assuming that the source port is pushing past 50% of that 1 Gps uplink.

That being said, I have a buddy who believes the C6K is fairly immune to this effect, on account of some aspect of its switching design.

Am I on the right track here, in terms of figuring out (a) how to assess potential impact to source traffic, and (b) in claiming that only the N5K has this SPAN rate-limiting command?

Before I go leaving sniffers SPANning 10Gb/s ports for long periods of time, I want to understand the risk of impacting production traffic.

--sk