TCP Sack option being removed on network layer between two EMC devices

[umccann1977]

Hi, this issue is fairly complex and I am interested to know if anyone else has run into this issue or has any recommendations where to focus our efforts in debugging this issue.

Issue:

EMC replication between two NS-120 devices is much slower then the WAN link should preform at.  In fact doing a simply CIFS transfer off between same EMC hardware over same network results in much better preformance on par with expectations.  EMC engineers believe the issue is due to the TCP Sack option being removed somewhere in the network layer between the two units resulting in data flow issues.

Basic Setup

Site 1: EMC NS-120 connected to 1GB interfaces on Cisco 3750 switches.  Traffic crosses IPSec Site-To-Site Tunnel over WAN link (20MB Internet connection) between Cisco ASA 5510 - 5540.

Site 2: EMC NS-120 connected to 10GB interfaces on Dell PowerConnect 8024 Switches

From what I understand TCP Sack option should be passed by default on all current Cisco switches / ASA equipment.  Is that a correct assumption. For the most part we have not modified the standard features on any of the switches or firewalls.  Would it potentially be removed in the IPSec Tunnel?

Any suggestions would be greatly appreciated.

Thanks

[susreeni · Accepted solution]

The EMC appliances in the topology appear to replicate fine if the bandwidth throttling is over 10MB. This setting was tried accidentally by the customer when doing some testing and a minimum bandwidth setting that the device supports is suspected. 

The customer is working with EMC on this behavior and believes that the Selective ACK is not being dropped and has nothing to do with the issue.