cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
685
Views
0
Helpful
2
Replies

RX drops with VNIC 1340

mdecker77
Level 1
Level 1

We are running an Oracle RAC database 2 node cluster on Cisco B200M4 / Firmware 3.2(3i)B / enic 2.3.0.53 /  and see RX drops for one VNIC interface. We then have increased RX/TX queues, ring sizes and RSS in Ethernet Policy and will enable MTU 9000 tomorrow. 

 

Still if I test with "iperf" and send packets with > 600 MBit/s over 10GbE interface, i see packet drops (ethtool -S eth1 -> rx_no_bufs).

 

Is this expected? Any idea on how to troubleshoot this further?

 

Regards,

 Martin

 

2 Replies 2

Kirk J
Cisco Employee
Cisco Employee

This generally indicates that the OS /Process in the OS isn't able to intake the packets fast enough.  RSS, RX ring Size buffers, Interrupts, are the commonly adjusted tunables for your scenario, which it appears you have already adjusted.

https://www.cisco.com/c/dam/en/us/products/collateral/interfaces-modules/unified-computing-system-adapters/vic-tuning-wp.pdf

Is the Baremetal OS, or a vitualized OS running RAC?

You running iperf with the -P and -w (-w 128M) switches?

[edit: use the -P to get parallel streams]

 

 

Kirk...

Hi Kirk,

thank you for your reply.

"-p" is for port. Yes, i have of course set a specific port.
"-w" is for window size/socket buffer size. No, I did not specify "-w".

However, we have specified these kernel parameters:

net.core.rmem_default=4194304
net.core.rmem_max=8388608
net.core.wmem_default=4194304
net.core.wmem_max=8388608
net.ipv4.ipfrag_high_thresh=16777216
net.ipv4.ipfrag_low_thresh=15728640


Regards,
Martin


Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: