Failover-Redundancy

[djanjic]

We are currently running hosted PBX on a physical Cent OS machine with approximately Our current load is somewhere between 30-40 concurrent calls at peak hours and 60 domains. We have now reached the point when this service is extremely critical and we simply have to have redundancy. Our plan is to move current configuration to a VM and provide redundancy by mirroring two machines in real time . Our question is what is recommended Hardware,( CPU cores, RAM etc). configuration for such a load provided that we are planning to grow to somewhere around 100 concurrent calls during peak hours.

 

Any suggestions-recommendations would be much appreciated.

 

Dusan

[Vodia PBX]

Not bring the biggest expert on this, my understanding is that the virtualization layer itself is not that expensive any more these days, as the CPU has special components for virtualization (e.g. see http://virtualization24x7.blogspot.com/2015/04/hardware-assisted-cpu-mmu-virtualization.html).Obviously you need to double check that your server supports those extensions, but my guess is that this pretty much mainstream these days.

 

One important thing about the PBX is to be able to provide CPU resources very fast, e.g. when a RTP is knocking on the door. The simplest approach is to have only one VM on a physical CPU, then there will be no doubt that if the PBX gets the CPU when it needs it. If you have this kind of load, it makes sense to reserve one CPU exclusively for the VM.

 

Also the network subsystem should be able to deal with virtualization. The server will have to listen to a bunch of MAC addresses. If the network adapter has to go into promiscuous mode, it means that every single packet will interrupt the CPU. Again without being an expert, I would assume that there must be some kind of hardware acceleration available also for the network side that triggers an event only if there is something really available. If you have 100 calls, there will be 10000 packets flooding the server per second; if that all has to be emulated in software it will drain some significant resources.

 

Maybe it makes sense to post the question also on a forum that deals only with virtualization.

[nebbin]

Do anyone have final answer on this redundancy approach. Would appreciate if you can share.

Thanks,

NE

[Vodia PBX]

Bottom line: Use a virtual machine. That will give you great redundancy within the data center.

Using the PBX failover makes only sense if you want to do a geographic failover ("nuclear strike"). Make backups in case you loose access to the VM, e.g. forgot to pay your bill.

[RichardDCG]

using the redundancy settings on the Vodia, when the secondary PBX instance detects a failover situation are the remote IP mappings a manual process so that tenants can reconnect?

[Vodia PBX]

16 minutes ago, RichardDCG said:

using the redundancy settings on the Vodia, when the secondary PBX instance detects a failover situation are the remote IP mappings a manual process so that tenants can reconnect?

If you want to do the Vodia failover, you need to think about DNS. One solution is to choose DNS SRV records, which require that all clients support that (which most do today). Another solution is to choose DNS A records and use the Action URL in the failover case to trigger a script which changes DNS records. Using IP address will not help unless your script if able to re-assign the IP address to the failover location. 

Reminder: In most (99 %) of the cases you don't need the Vodia failover because your datacenter already provides you with failover of the VM without you having to do anything. The Vodia failover is for geographical failover and some really special cases, and it is a huge amount of work to set this up properly (and not make things worse). 

[RichardDCG]

there is some basic redundancy that comes standard with EC2 instances in AWS. I guess the PBX becomes more and more critical as we increase the number of tenants and the potential for 24/7 access is required then we need to look at additional redundancy on top of that.  Potentially just having periodic AMI images created may suffice...

[Vodia PBX]

3 hours ago, RichardDCG said:

Potentially just having periodic AMI images created may suffice...

Absolutely. Failover when the datacenter is under attack and goes down is one thing. But when an administrator accidentally makes a stupid move is another. I would say the latter happens more often. Then having a snapshot not too long ago can be a true blessing. And it's really easy and cheap in most data centers, including EC2.