IBM Support

Network performance tuning using flip_n_run

Education


Abstract

AIX latest network adapters has an attribute flip_n_run. The value of the attribute shall be set based on the workload to get maximum performance.
This article describes more details on the attribute.

Content

When a network packet is received through a network adapter, it is sent to driver and hence the upper layers. 
In TCP, if any send happens as part of processing the received packet, the send shall happen through the same 
thread context through which the packet is received. Such cases may delay the processing of further packets received 
waiting in the Work Queue of the adapter. This may cause some performance impact on the network. This is applicable
for high speed adapters like 100G.
To avoid this users shall enable the attribute flip_n_run on the adapter to yes, which causes the packets sent to be queued 
in a software queue of the driver, and sent offlevel through a timer. 

[{"Business Unit":{"code":"BU058","label":"IBM Infrastructure w\/TPS"},"Product":{"code":"SWG10","label":"AIX"},"Component":"","Platform":[{"code":"PF002","label":"AIX"}],"Version":"All Versions","Edition":"","Line of Business":{"code":"LOB08","label":"Cognitive Systems"}}]

Document Information

Modified date:
16 September 2019

UID

ibm10888059