Cisco UCS 6332 16UP Performance Issue

[Xdxb]

Hi,

Does Anyone face performance issue due to hardware issue on Cisco UCS 6332 16UP?

I am facing: Whenever Network utilization spikes on 40g ports, reports high input discards.

Please share your experience. 

[Kirk J]

The 6454s have a couple of cosmetic bugs where the input discards are not real.  Not sure if the 6332's (different ASIC) have similar issue in code.

Are these server ports or uplink ports?

Assuming you have confirmed the corresponding port programming on other side of the link has same allowed VLANs as I think that can also contribute to drops(typically broadcasts for VLAN not allowed on FI port).

What is your 'actual' performance impact?  Are you seeing a lot of retransmits or something?

Kirk...

[Steven Tardy]

To add to what Kirk has said.

If you have mismatched speeds like:

• Servers connected at 10Gbps and Uplinks connected at 40Gbps
• Servers connected at 40Gbps and Uplinks connected at 10Gbps

Then by definition you have over-subscription. Over-subscription requires buffering. When the buffers are full there will be discards.

UCS QoS can sometimes increase discards by enabling (unused) QoS classes thereby taking some buffers away from the default buffer.

Is this a "real" issue or a "perceived" issue? First thing I would do is take a packet capture to know if the servers/endpoints are seeing discards (and TCP retransmission) or is this counter untruthful.

[Xdxb]

Very beginning, we had connected 10g Servers to FI, but uplink was 40G. Later on we added 40g Servers to same FI.

• Servers connected at 10Gbps and Uplinks connected at 40Gbps
• Servers connected at 40Gbps and Uplinks connected at 40Gbps

Whenever we perform any activity like vMotion or storage vMotion, copying big data reports high input discards on uplinks