Network Redundancy for Master/Media Servers

[Alex_Vasquez]

In our Netbackup Environment we have 1 master server and 15 media servers. All of these servers are connected to one network switch; we also do not cluster. We're running NBU 7.0.1. We want to make our media servers reduntant on the network in case we lose a switch we can continue backups. Also, if there's a way to automate we'd like to explore that direction as well.

My question is this: What is the appropriate configuration for Plugging in Netbackup Media/Master servers to two different switches? Is there a way to automate? Or is there only manual config to get things working appropriately?

You advice is appreciated.

[AAlmroth]

If your LAN infrastructure can do it, then you could potentially create a team/bond/aggregate of two NIC ports on your media servers, connected to separate switches. If the switches/core can load-balance/route/switch the traffic between the two LAN fabrics then you have the simplest solution, using only one IP address.

This is a very common setup, so most infrastructures can at least handle failover teams. The more interesting part is when you can do LACP over two LAN fabrics, but not all network admins like this, as it potentially put a lot of stress on the inter-fabric links.

There is no automatic process, but manually create the team on the host, using two NICs connected to two switches, and in the case of LACP, configure the switch/es accordingly.

The other alternative is to use two IP addresses, and by design "load-balance" the client backups using the two different IP addresses. This was the model we had to use before teamed NIC ports on separate LAN fabrics was possible... I wouldn't recommend this approach today.

/A

[Alex_Vasquez]

Thanks for the reply, AAlmroth.

You've given me some good things to think about. Now, if we team the nics on our media servers, would that potentially have negative impact on Netbackup? I seem to recall reading of instances where people had trouble with the setup. I'm just not sure if there's a best practices doc for this type of setup. Do you know of anything?

[AAlmroth]

I think problems with NIC teaming is something we saw in the "early days" where the quality of the teaming software was not up to the task.

A common problem back in the early days was that spurious load on the NIC team was handled fine, but sustained high load (such as in backup applications) could not be handled, and thus interruptions in traffic (which NBU is quite unhappy with).

My experience with NIC teaming is that Intel based NICs works much better than Broadcom on the Windows platform. On Linux, where a general bond driver is used, the type of NIC is less important, as well as the Intel and Broadcom drivers usually comes with the OS distribution.

The other factor with teaming is that we cannot expect linear performance, that is if using 4 x 1Gbit NIC, we will not achieve 4Gbit/s performance, but rather 80%. This is due to TCP/IP overhead as well as the link aggregation protocol (LACP).

For best backup performance, the switches should be configured to use LACP for the ports used by the media server, and have the team on the host set to use LACP. LACP is also referenced in IEEE802.3ad and IEEE802.1AX standards. This will allowed maximum input to the media server, which we want during backups.

/A

[Zahid_Haseeb]

I think it seems that you should cluster the switches. Discuss this matter with your Network Infrastructure team.

I am attaching a link below with my post which may clear your concept i hope.

Managing Clusters of Switches

http://www.cisco.com/en/US/docs/switches/lan/catalyst2900xl_3500xl/releasesa6/scg/kiclust.html

[Alex_Vasquez]

Zahid,

Thanks for the useful post! I think I can use this get some good information to my network team. =)

Thanks again.

[Alex_Vasquez]

This is good information! Thanks again for taking the time to give your input. I really do appreciate it.

At least I can begin the process of discussion on this matter with my network team. Thanks again!