Tcpdump for IP Metrocluster fabric interfaces

[Ofp]

Hello.

I have got somewhat strange question.

Exist AFF-A800 IP Metrocluster with Juniper switches as cluster network. Why Juniper? Because Cisco switches aren't available in next couple months.

The metrocluster has problem: When write-traffic periodically increase - latency dramatically jump. Our network engineers can't exactly translate settings for Cisco switches to settings for Juniper switches because they have a different set of settings for traffic priority.

I suppose that we can chance to find root of problem if we collect traffic dump for investigation. Question: How I can collect tcpdump on ip metrocluster fabric interfaces (e0b, e1b)? These interfaces are hidden in ONTAP.

[TMACMD]

Did you happen to read this:

https://docs.netapp.com/us-en/ontap-metrocluster/install-ip/concept_considerations_mc_compliant_switches.html

Specifically in that page:

Switch and cabling requirements

• The switches must support QoS/traffic classification.

• The switches must support explicit congestion notification (ECN).

• The switches must support L4 port-vlan load-balancing policies to preserve order along the path.

• The switches must support L2 Flow Control (L2FC).

• The cables connecting the nodes to the switches must be purchased from NetApp. The cables we provide must be supported by the switch vendor.

Do your Juniper switches meet all those requirements?

The hidden ports do mostly iSCSI traffic. 

[Ofp]

Yes.

Our network engineers checked this documentation and answered - "Yes. Our switches Juniper QFX5120 are meet these requirements".

[paul_stejskal]

It may be possible from the BSD layer, but I would recommend opening a case with Support. Make sure you tag this as a Metrocluster issue. If you mention slowness it will go to us in the Perf team and we will have to engage MCC folks. At the very least we could see if we could run tcpdump for those interfaces.

[Ofp]

Of course we made a case.

[ttran]

Hello @Ofp ,

I am unable to find any documentation that we have tested the specific model switch Juniper QFX5120 for any of our configurations. We are more strict as to which switch vendors and models are capable and supported for Cluster communication compared to client-access switches. 

Please continue to work this issue through the Support case and/or your Account Team as those are the correct avenues to get the Juniper QFX5120 working.

Regards,

Team NetApp

[tahmad]

Hello @Ofp ,

Did you open a case with support? What was the recommendation?