High availability with Cluster on two servers via wan

[Rocha Neto]

Hello friends okay?

I wonder if dear friends could get me some doubts as to the Cluster with high availability.

I have 2 servers and I would like to make the cluster of them. But I have a few questions.

1 server is in a rack 11 and another server in a rack 13. I can't call them directly. The datacenter informs you that I can't make the call because they're in different racks.

1 server has the track 192.168.88.33/27
2 server has the track 192.168.68.65/27

How can I make the link between these two servers with high availability?

They only communicate by wan.

I wonder if I can have a good performance?

Thank you.

 

[Macavity]

What does "i cant call them directly" mean? You can not send a ping from one host to the other?

As both IPs are on a private subnet the speed will pedent on the switch / newtwork connection which is used between them.

 

[Rocha Neto]

Hello thanks for your response.

I can't do the direct communication between local network. I can drip the ips without any problem. But they do not communicate across the LAN, WAN only.

That is the request comes out and goes, rather than get into the router and go to the destination server.

My case is as if the servers were in different locations.

 

[MimCom]

Please correct me if I'm wrong, but 'dripping' an IP seems like a translation problem.

Can you ping from each of those IP's to the other? If not, you may need to create a tunnel between the sites that will route those prefixes.

 

[Rocha Neto]

Sorry, it was a translation error.

I am able to ping between the IPs.

The servers have normal communication over the internet WAN.

 

[MimCom]

What is the WAN comprised of? Is it a private link between two datacenters or is there public Internet involved?

Are there firewalls in place on either end?

 

[Rocha Neto]

Only Public Internet.

 

[Macavity]

But both IP you namend in your first post are from a private IP Range which is not routed over the Public Internet. Or do they have been just examples?

 

[Rocha Neto]

Are only examples. To show that they are not in the same range or same network

 

[Ashley]

Your best bet is to check the WIKI on the multi casting debug/testing.

Run some tests and see if multicasting will even work between the two servers, I would be surprised if it's enabled to be honest on a public shared network.

 

[MimCom]

I was not aware that the addresses provided were placeholders.

As I previously suggested, a tunnel might be the best way to handle this. What kind of bandwidth do you have available on these connections, and how much of that do you want to use for the HA traffic?

 

[Rocha Neto]

Hello

I have 30 TB on each server. Could you talk more about the tunnel? Is there any material where I can study on this subject?

Thank you.

 

[Ashley]

What storage will you be using on the HA setup?

 

[Rocha Neto]

In a node, I use LVM-Thin, I have 4 x 1 TB Raid 10 HW (each server). I haven't thought about it. But should I use NFS.

 

[Ashley]

For HA your really need a shared storage to make it any sense to do.

And as you don't have any private network between the two servers it makes any form of shared storage pretty unrealistic.

Have you checked with your DC about a seperate private network?

 

[MimCom]

30 TB is a fair bit of data, but the bandwidth required really depends on how dynamic it is. What bandwidth (speed) connections are available at each end? Have you tried running iperf between the nodes (both directions)?

By tunnel I was implying something like VPLS (assuming the intervening networks will support that) or perhaps ipsec with hardware acceleration.

 

[Rocha Neto]

30TB on 1Gbps Bandwidth on each server.

Haven't ran the iperf between nodes. I'm going to do that.

Thank you for your interest in this topic. Are helping.

 

[MimCom]

Make sure you option multiple streams (-P.) iperf3 is a bit less demanding on the nodes.

 

[Rocha Neto]

I talked to the Datacenter.

They replied: This is not currently the function of the network but it is in the roadmap to implement. We are sorry for the inconvenience this may cause you. Please let us know if you need anything else and we will be glad to assist you.

 

[MimCom]

By 'This' do you mean VPLS?

If so, and you do have good speed/latency between the centers, you could still use something like an IPsec tunnel in the meantime.