Cisco 3750x Connected to Netapp FAS2040 showing "OutDiscard" Errors - Flow Control?

[richard_pery]

Hi all,

I am after a bit of help in regards to a recently installed NetApp FAS2040.  Connectivity is as follows:

• 1 x iSCSI interface connected to VLAN10 on Cisco 3750
  
• 1 x iSCSI interface connected to VLAN20 on Cisco 3750
  
• 4 x HP DL380 G7 servers with 1 x HP NIC into each VLAN
  

All links are running at 1000 Full Duplex.

I am getting a few "OutDiscards" appear on the two Cisco ports connected to the NetApp.  Initial investigation said about duplex / speed issues, but all appears to be correct.  No errors are showing on the HP Server NIC's, nor are any other error counters showing hits.

I have checked using 'ifstat -a' and none of the NetApp interfaces have 'flow control' enabled, nor have I done any config on the Cisco switches.

My question (finally...) is would it be worth turning on Flow Control on the NetApp and the Cisco switches, I have read various docs / forums (not NetApp specific), and can't seem to get a definate yes / no answer.

Any help / pointers would be helpful!

Kind regards,

Richard.

[kevingraham]

richard.pery wrote:
I am getting a few "OutDiscards" appear on the two Cisco ports connected to the NetApp.  Initial investigation said about duplex / speed issues, but all appears to be correct.  No errors are showing on the HP Server NIC's, nor are any other error counters showing hits.

This is almost certainly an artifact of the small output buffers on the 3750 and microbursts. Even on a sub-second basis, the fan-in from the iSCSI hosts exceeds the rate of the interface to the filer. That forces the switch to queue packets until traffic lets up enough for them to be transmitted (this is the same behavior as when a host sends XON/XOFF flow control -- it just triggers that queuing prematurely so the host can have time to consume traffic off the wire). When that queue is exceeded (I believe its <100ms for a gigabit interface on this platform) packets are dropped, which is what you're seeing.

You can see these with 'show platform port-asic stats drop'. If you're seeing them _without_ flow-control enabled (since it can trigger this prematurely), then this is definitely the issue.

[richard_pery]

Hi,

Thanks every so much for your response, this does make sense.  It only seems to be a few errors being generated every day.  We are aiming to add further physical servers next year, so I guess the errors may increase.

I have ran the command 'show platform port-asic stats drop' it doesn't show any errors, running the command 'show platform port-asic stats drop gigabitEthernet 1/0/11' gives the following info:

************************************

Queue 3
      Weight 0 Frames 0
      Weight 1 Frames 0
      Weight 2 Frames 6310

************************************

Would either of the following help:

1) turning on jumbo packets (will need upgrade of Citrix ZenServer farm from 5.6 to 6 for full support)

2) Enabling flow-control fully in iSCSI interfaces

Kind regards,

Richard.