Passport 1648T Network Issues: Windowing, Buffer ?

[alivedata]

Hello all,

I have a very interesting problem. We have a Passport 1648T in production with 4 VLANS including the default. It seem to work very well, in most circumstances, however, during backups, server to server, it tends to drop the connection. This does not happen on backups that connect via unc, or crossover but rather backups that utilize an agent via the switch. From my understanding an agent is told what information to transmit and the backup server waits for and writes the data. The switch seems to be the issue. Firmware is up to date and settings, MTU etc, are default. I believe that the switch is dropping the connection at one point but can't pinpoint why/where. Could this be an issue with windowing? A buffer problem? CPU utilization? A specified timeout period? I am lost, and this only happens with Veritas Backup Exec agent for windows servers and robocopy...network name cannot be found

Thank you in advance

 

[Jynxx]

I would not be suprised if it is a similar buffering issue that is notorious on the 5510's.

Likely the TCP window size is rather large for the backup programs.

On the 5510's, they would choke when transferring from one port speed to another, high to low. For instance when transferring from 1000 to 100, or from 100 to 10.

This would always occur with anything higher than about 20000 for the TCP window size.

This was resolved with 4.0 code on the 5510's, as well as specifying a "max buffer size" as well as a queue-set with no more than 4 queues used to divvy up buffer space for QoS.