We usually see the TCP zero window from NS SNIP to backend server IP and this sometimes indicates that the NetScaler does not have sufficient space in buffer to process TCP traffic quickly enough.



To alleviate this issue, the following KB suggests to change the default TCP profile settings:

• TCP flavor as BIC and
• Increase the TCP (receive) buffer size.

Retransmission Timeout Causes Network Latency on SSL Connections through NetScaler 
https://support.citrix.com/article/CTX205656
 

From the past cases, we recommend this value to be set to 128KB~256KB
If we turn this too big, for example 2MB~4MB, although we can significantly reduce the TCP zero window and TCP retransmission there will be more connections in surge queue leading to latency.

Note: The configuration change example below is applied to the default profile which is applicable to most scenarios. If there is a custom profile in place and bound to an entity, then changes need to be done on that profile as well.







 

This article discusses increasing TCP buffer size to alleviate NetScaler zero window issue