How to increase Keepalive and TcpMaxDataRetransmissions to solve intermitte

Question

Hi experts,
  I encountered an intermittent link down issue on our production system for a long time. I escalated to Cisco TAC but the problem is too difficult to resolve(no errors in Cisco Nexus 5010). We built a very solid HA in network(4 * 10GBE teaming), 2 Cisco Nexus 5010 for HA. However, intermittent link down occurred from time to time. By the way, chimney offload is disabled recently but the problem persists. We run SQL server 2008 R2 CU7 on Windows 2008 R2 SP1. The NICs are HP/Emulex NC550SFP 10GBE.
  I think this problem can't be solved by Cisco so I want to increase Keepalive from 30,000 to 300,000 on DB server and TcpMaxDataRetransmissions from 5 to 120 on DB/APP servers.(according to this link: http://blogs.msdn.com/b/dataaccesstechnologies/archive/2010/06/18/using-network-trace-to-troubleshoot-intermittent-connectivity-issues.aspx ) However, I test these settings in testing environment and run a query in query analyzer and I found it does not work. Query analyzer terminated a query within 5 seconds if I unplug the cable. What's wrong with that?

Answer 1

Hi Dennis,
Keepalive property is a new stuff set on SQL server configuration manager from SQL 2005.
This parameter controls how often TCP attempts to verify that an idle connection is still intact by sending a keep alive packet to its peer. If the remote system is still reachable and functioning, a acknowledge packet is sent back. Otherwise, the local TCP will keep sending u201CKeep Aliveu201D packet in an interval of KeepAliveInterval for TcpMaxDataRetransmissions times. If no acknowledge packet is received during this period, the local TCP will reset the connection. For every SQL Server TCP connections.
To cut it short , for an idle connection once in every 30000 milli sec a check is sent to the client and check whether the client is available . If yes , the connection is maintained else the conenction is killed / reset.
Based on this you and your manager can decide whether or not this property is to be set on .
To play around with this property you can do it from here
SQL Server Configuration Manager -> SQL Server 2005 Network Configuration ->Protocols -> TCP/IP, you will notice the newly added property u201CKeep Aliveu201D.
if you have SQL server 2008, it is under SQL Native Client > Client Protocols > TCP/IP
Additionally you can have a look into a older blog
http://blogs.msdn.com/b/sql_protocols/archive/2006/03/09/546852.aspx
Hope this helps.
Regards,
Deepak Kori