Unsolved
This post is more than 5 years old
3 Posts
0
33731
July 29th, 2008 12:00
Setting up two 6248 in a High Availability setup (2 NIC servers attached to each switch)
I am okay with setting up servers, but with network equipment - not so much :)
I have Linux servers with 2 NIC cards. I would like to take 2 x 6248 and connect (stack?) then together.
I would then like to attach the Linux servers so that one NIC card goes to each switch.
In case of a switch failure, everything will still be able to talk to each other.
"Bonding" the NICs on the Linux side is easy, but what do I have to do on the switch side so that the switches communicate with each other and know how to route everything correctly?
The switchs are just the factory default settings (ip, snmp, etc). There are no VLANS, etc. It is a simple a setup as possible.
TIA!
No Events found!
magnus.tengmo
1 Rookie
•
9 Posts
0
July 29th, 2008 12:00
I got the same question, to stack or not to stack. If I understand userguide, if you stack switches, masters need to be up, if master reboots (example, during firmware upgrade) all ports will be down for a short time.
In your case, I think you should use a channel-group between both switches.
I found some nice whitepapers on this link:
http://www1.ap.dell.com/content/topics/global.aspx/solutions/en/pwcnt_papers?c=th&l=en&s=lca
Best Regards, Magnus
Alex_Spitzer
3 Posts
0
July 29th, 2008 14:00
Right, I am not trying to extend the switch from 48 - 96 ports (so maybe stacking is the wrong word.)
I want 48 "redundant" ports.
If one switch dies, and that effects everything, then I might as well just use one switch.
Do I just enable spanning tree, and run a cable between the 2 switches (any port?), and nothing else? That is sort of what it looks like they are saying in the link labeled
"High-Availability Networks with Spanning Tree and Rapid Spanning Tree Protocol"
on the url you listed.
magnus.tengmo
1 Rookie
•
9 Posts
0
July 29th, 2008 14:00
Spanningtree is for detect loops and build a redundant network between switches. Good, but could be complex to setup (mstp with loadbalance different vlan). HP have a good guide how SPT works: http://www.hp.com/rnd/training/technical/MSTP.htm
If you only use two switches, i should have create a channel-group between two switches with 2 or 4 ports to get speed and redundancy between the two switches.
See this document: http://www.dell.com/downloads/global/products/pwcnt/en/pwcnt_link_aggregation.pdf
I don´t think you need to change anything on server ports.
Good luck, Magnus
Alex_Spitzer
3 Posts
0
July 30th, 2008 12:00
Is Spanning Tree complex to setup? I thought I just "turned it on"...but again, I am not a network guy.
also a bit more info on what I want to do:
The "core clustered switches" would have all the databases (dual attached) which all our other servers use. These should be accessable all the time, hence the idea of making them redundant.
In seperate cabinets would be single "leaf" switches that all the servers in that one cabinet plug into. These single switches would need 2 cables running back to the "core clustered switches" (one to each of the "clustered switches")
In our environment, it is okay to lose a cabinet of "drone" servers (due to the "leaf switch" dying) as it just means that we have 25 less servers processing data until it is fixed. We would not like to lose switches that the databases are plugged into though, as that means ALL the servers would stop processing until THAT switch is replaced.
sentinel-master
345 Posts
0
August 3rd, 2008 17:00
I think what you are really asking for is "redundant NIC" operation for the server - so high availability is maintained, should the "primary NIC or switch" fail. This concept is also utilised in VMware virtualisation concepts.
So essentially each NIC (fault tolerant mode) goes to a different "access switch - 6248", which has implications on the network topology. Because you want to maintain the same IP address from the failed NIC to the redundant active NIC, your now need to be in the same VLAN (subnet) , when failing across or activating the redundant NIC.
Hence both switches must have the same VLAN or set of VLANs present - or fundamentally this will not work. This requires those VLANs to be omni present across the switch infrastructure in a classical triangle shape from Distribution Switch A to Leaf switch C to Distribution switch B. Because a loop or triangle exists you need spanning tree to block the loop, and be able to recover or change the topology in the event of a switch or link failure automatically.
If your networking people are familiar with spanning tree or rapid spanning tree - then yes, you do simply need to turn it on - on all switches. The key point to ensure though is set spanning tree root for each VLAN and also set "Next designated root", this ensures in the event of a major switch failure or someone powering off "ROOT" switch, the network topology goes to a predictable & working state.
There are a couple of further enhancements to high availability designs for servers, let me know if this makes sense first.