We're thinking of buying a MFSYS25 chasis blade with two switches AXXSW1GB (adding extra mezzdadines) but we don´t know what is the best way to have fault tolerance, maximum bandwith and high availability.
Our platform will use Linux redHat and perhaps KVM instead of vmware for virtual machines.
We've been reading about Fault tolerance feature and teaming in switch (http://www.intel.com/support/motherboards/server/mfsys25/sb/CS-028603.htm Intel� Modular Server System MFSYS25 — FAQs about the Intel� Gigabit Ethernet Switch AXXSW1GB) but it's not so clear and perhaps it could be better way making a linux bonding between one nic of one switch and one nic of other switch and cross connect both switches. The issue is we would need one port of each switch per VLAN and perhaps we couldn't make LAG/LACP between both switches. Of course, I suppose that spanning tree should be disable.
Could you please help us about what could be the best way? Teaming NICs by Hw? Bonding NICs by Linux? If we bond using Linux or KVM, could we bond these four NICs? And such case, could we make LAG between ports of these two switches for cross-connect purposes?
Thank you in advance.
The two switches are not physically connected to each other inside the chassis, so just think of them as two separate switches, each connected to two NIC ports on the compute blade. If you want fault tolerance between the two switches, use Switch Fault Tolerance (SFT) to team NIC1 or NIC2, with NIC3 or NIC4.
For more information about SFT, see http://www.intel.com/support/network/sb/cs-009747.htm Network Connectivity — How do I use Teaming with Advanced Networking Services (ANS)?
Thank you for your answer. So, it seems that the best way is making a teaming between one NIC of one switch and another NIC of the other switch and cross connect both switches. Isn't it?
What about aggregating ports? Is it possible to LAG more than one port for cross-connect purposes?
Yes you're correct.
When you say aggregating ports, do you mean the connection between the switches? Yes you can configure LAG for the uplink switch ports.
Yes,what we want is to aggretate more than one port (uplink) between both switches. Please, follow next images. Suppose that one switch has NIC 1 and 2 and the other switch has NIC 3 and 4. Firewall is defined as master/slave, so it can change from one to another at any time. The idea is making a linux bonding (or vmware teaming) between NICs and maximize the 10 uplinks port of each switch:
On this image, what I try to show is a teaming between NIC 1 and 3 (VLAN 1) and NIC 2 and 4 (VLAN 2). The issue on this topology is we should use one port for each switch and for each VLAN for cross-connect them. For this reason, perhaps it could be better choice next image:
Here, we have a teaming of four NICs and a LAG of two (or more) ports between both switches. Is it possible to aggregate these two uplink ports with VLAN tagging?
Yes. You can configure LAG first, and then add the LAG to the VLAN. For more details, see the http://www.intel.com/support/motherboards/server/mfsys25/sb/CS-028550.htm AXXSW1GB user guide chapter 5 and 6.