Hi
I was wondering how to achieve an architecture with two or more active load-balancer nodes (kamailio+dispatcher module).
I have read how to setup two dispatcher nodes, one node as a master and the other one as a backup.
The backup node doesn't process any traffic until master fails. But how to make the traffic being processed by both load-balancers ? In the case for instance the capacity of single node is not enough to process all incoming traffic. Is there any recommended configuration (eg. using frontal application-level swith)
Regards, Pascal
Any "frontal application-level switch" would simply have the same liability.
Kamailio is a very high-throughput proxy that can handle huge amounts of call setups per second. I think you can count on that.
Failing over around the load balancer node to a secondary load balancer or distributing the traffic among multiple load balancers is a job best left to the sending endpoint. For example, a DID origination provider's switch or SBC can be set up to fail over calls to a different IP endpoint for your SIP trunk if no response is received within a certain amount of time. That is how this is typically done. At some point you've got to say that you've done all you can do, and it's up to the other side.
Pascal Maugeri wrote:
Hi
I was wondering how to achieve an architecture with two or more active load-balancer nodes (kamailio+dispatcher module).
I have read how to setup two dispatcher nodes, one node as a master and the other one as a backup.
The backup node doesn't process any traffic until master fails. But how to make the traffic being processed by both load-balancers ? In the case for instance the capacity of single node is not enough to process all incoming traffic. Is there any recommended configuration (eg. using frontal application-level swith)
Regards, Pascal
Users mailing list Users@lists.kamailio.org http://lists.kamailio.org/cgi-bin/mailman/listinfo/users
Hello,
a solution I am using is to do shared IP (vrrp/ucarp). The typical scenario is active-backup, when one share IP is used.
But you can do active-active by using two shared IP addresses, so will be
server1: active_ip_1 - backup_ip_2 server2: backup_ip_1 - active_ip_2
So each server is active and backs-up the other. This can be scaled, with more than 2 servers, you need an IP per server, and you will have a chain of active servers backing up next one. Based on who is backing up who, you may have one or more servers down without affecting the service
You have to take care of balancing the traffic among the IP addresses.
Cheers, Daniel
On 10/23/08 13:16, Alex Balashov wrote:
Any "frontal application-level switch" would simply have the same liability.
Kamailio is a very high-throughput proxy that can handle huge amounts of call setups per second. I think you can count on that.
Failing over around the load balancer node to a secondary load balancer or distributing the traffic among multiple load balancers is a job best left to the sending endpoint. For example, a DID origination provider's switch or SBC can be set up to fail over calls to a different IP endpoint for your SIP trunk if no response is received within a certain amount of time. That is how this is typically done. At some point you've got to say that you've done all you can do, and it's up to the other side.
Pascal Maugeri wrote:
Hi
I was wondering how to achieve an architecture with two or more active load-balancer nodes (kamailio+dispatcher module).
I have read how to setup two dispatcher nodes, one node as a master and the other one as a backup.
The backup node doesn't process any traffic until master fails. But how to make the traffic being processed by both load-balancers ? In the case for instance the capacity of single node is not enough to process all incoming traffic. Is there any recommended configuration (eg. using frontal application-level swith)
Regards, Pascal
Users mailing list Users@lists.kamailio.org http://lists.kamailio.org/cgi-bin/mailman/listinfo/users
On Thu, Oct 23, 2008 at 12:16 PM, Alex Balashov abalashov@evaristesys.comwrote:
Any "frontal application-level switch" would simply have the same liability.
right.
Kamailio is a very high-throughput proxy that can handle huge amounts of call setups per second. I think you can count on that.
Yes but you may need to use the power of all your load-balancer nodes. I believe we can expect 10K transactions/seconds on a single load-balancer node, but if you need to handle more traffic, load balancer node may be a bottleneck.
Failing over around the load balancer node to a secondary load balancer or distributing the traffic among multiple load balancers is a job best left to the sending endpoint. For example, a DID origination provider's switch or SBC can be set up to fail over calls to a different IP endpoint for your SIP trunk if no response is received within a certain amount of time. That is how this is typically done. At some point you've got to say that you've done all you can do, and it's up to the other side.
Thanks Alex, it's a good point.
-pascal
Pascal Maugeri wrote:
Hi
I was wondering how to achieve an architecture with two or more active load-balancer nodes (kamailio+dispatcher module).
I have read how to setup two dispatcher nodes, one node as a master and the other one as a backup.
The backup node doesn't process any traffic until master fails. But how to make the traffic being processed by both load-balancers ? In the case for instance the capacity of single node is not enough to process all incoming traffic. Is there any recommended configuration (eg. using frontal application-level swith)
Regards, Pascal
Users mailing list Users@lists.kamailio.org http://lists.kamailio.org/cgi-bin/mailman/listinfo/users
-- Alex Balashov Evariste Systems Web : http://www.evaristesys.com/ Tel : (+1) (678) 954-0670 Direct : (+1) (678) 954-0671 Mobile : (+1) (706) 338-8599
On Thu, Oct 23, 2008 at 8:25 AM, Pascal Maugeri pascal.maugeri@gmail.com wrote:
On Thu, Oct 23, 2008 at 12:16 PM, Alex Balashov abalashov@evaristesys.com wrote:
Any "frontal application-level switch" would simply have the same liability.
right.
Kamailio is a very high-throughput proxy that can handle huge amounts of call setups per second. I think you can count on that.
Yes but you may need to use the power of all your load-balancer nodes. I believe we can expect 10K transactions/seconds on a single load-balancer node, but if you need to handle more traffic, load balancer node may be a bottleneck.
I am doing between 4-5K simultaneous transactions (100-150 calls per second) on a Xeon quad core @3.20GHz and the overal cpu load is under 5%. I don't think that your load-balancer node will be a bottleneck. You just need a properly tunned config file.
Failing over around the load balancer node to a secondary load balancer or distributing the traffic among multiple load balancers is a job best left to the sending endpoint. For example, a DID origination provider's switch or SBC can be set up to fail over calls to a different IP endpoint for your SIP trunk if no response is received within a certain amount of time. That is how this is typically done. At some point you've got to say that you've done all you can do, and it's up to the other side.
Thanks Alex, it's a good point.
-pascal
Pascal Maugeri wrote:
Hi
I was wondering how to achieve an architecture with two or more active load-balancer nodes (kamailio+dispatcher module).
I have read how to setup two dispatcher nodes, one node as a master and the other one as a backup.
The backup node doesn't process any traffic until master fails. But how to make the traffic being processed by both load-balancers ? In the case for instance the capacity of single node is not enough to process all incoming traffic. Is there any recommended configuration (eg. using frontal application-level swith)
Regards, Pascal
Users mailing list Users@lists.kamailio.org http://lists.kamailio.org/cgi-bin/mailman/listinfo/users
-- Alex Balashov Evariste Systems Web : http://www.evaristesys.com/ Tel : (+1) (678) 954-0670 Direct : (+1) (678) 954-0671 Mobile : (+1) (706) 338-8599
Users mailing list Users@lists.kamailio.org http://lists.kamailio.org/cgi-bin/mailman/listinfo/users
-
Alex Balashov -
Daniel-Constantin Mierla -
Ovidiu Sas -
Pascal Maugeri