Network SIP Server/ SIP Server HA / Load-Balancing

[victor]

Hi, guys,

I am having some very big doubts about one of our designs here, and I was hoping someone could shed some light on how to address them.

First of all, a small overview:

we have two sites, with 1000 agents each. Both sites are using Genesys SIP Server with NET.COM's ShoutIP as a VoIP gateway. Should Site A get nuked, operators using site A just connect to Site B and continue their business. (regardless of nuclear winter, for which each of them is provided with a very warm fur coat)

Straight away I have some serious problems with the design:

Load-Balancing
SIP T-Server is designed for only 1000 agents at a time. Since we have 1000 agents per site, I am uneasy taking Genesys at its word about the capacity and would love to break it down to 500 agents per server. Unfortunately, if I create SIP T-Server Site A_1 and SIP T-Server Site A_2 then it means I will have four sites altogether and not two: Site A_1, Site A_2, Site B_1, Site B_2. Obviously, we do not want to do that.
Can I assign TWO SIP T-Servers to one site?

Network SIP Server seems to be an LDS equivalent for SIP Servers, distributing load between different SIP T-Servers, but unless it is possible to assign TWO or MORE SIP T-Servers to the same site (Switch definition in CME) and those SIP Servers would synchronize information between them (read: HA), I am not sure how we can use it.

High Availability
I have major problem with the way Genesys SIP T-Server is designed in regards to HA. They want you to use some sort of Network Load Balancer such as BigIP or WCS, which means that:
- there is only one virtual IP and unlike other products it is impossible to handle switchover in realtime without any adverse affects
- it is not designed with load-balancing in mind.

So, my questions in short are:

1. is it possible to assign more than one SIP T-Server to the same switch and load-balance between them?
2. if we are to use Network SIP T-Server, is the only way to get HA and LB is to WCS all SIP-Tservers in pairs of two?

My dream setup would be similar to the way T-Server and URS HA is setup:
- you have complete HA and LB implementation without the need for any WCS.

How do I do it?

Thanks!!!
Vic

[victor]

Seriously, anyone?

[Timur Karimov]

hi!

Victor, as i know u can't do it. u can't assign two sip-server to one switch. only one sip-server and it ha-pair.

[Fra]

Vic,

I wouldn't ever try to assign two T-Servers to the same CME switch; in case of SIP-Servers, maybe you are able to do that and maybe each of their "T-Server" entity will register only the set of DN's which have been "SIP-registered". But the main issue you would face is about internal routing, cause you wouldn't know where the called endpoint is registered to...
In my opinion there is no way to get two SIP-Servers synchronize, unless they are a HA pair.
As far as I know, Genesys state that the only way to configure high availability is to use Windows NLB (Unix Virtual IP).
I know you would'nt like that, but the solution that would work would be having a HA pair of Network SIP-Servers load-balancing four HA-pairs of SIP-Servers 

Fra

[msassoon]

That is correct. Also, the need for NLB or Virtual IP is a restriction of the SIP protocol. Normally, SIP endpoints do not allow for a primary and secondary address. The gateways will probably be fine, but the problem is usually the SIP phones used by the agents. You could use a Load Balancing server, but REGISTER messages will be the issue.

[Timur Karimov]

[quote author=Fra link=topic=3786.msg16281#msg16281 date=1233315354]
Vic,
I know you would'nt like that, but the solution that would work would be having a HA pair of Network SIP-Servers load-balancing four HA-pairs of SIP-Servers 

Fra
[/quote]

it's work. but what i don't undestud - why we need stupid script for up and down interface on a second sip-server?

[victor]

msassoon: I always thought that SIP does support redundancy protocol. I have read several RFCs to this effect... Am I wrong? [url=http://tools.ietf.org/html/draft-worley-redundancy-response-00]http://tools.ietf.org/html/draft-worley-redundancy-response-00[/url]

As for end-points, most of SIP stacks should be able to handle this... I think. At least softphone ones. And with Genesys SIP softphone sample using RTC, that should not be a problem either...I think 

Even if redundancy was not suppoted by SIP protocol, I am sure Genesys could have modified their LDS to act as a proxy between Softphone and SIP Server and thus get rid of the annoying NLB, which costs money and does a very poor job of synchronizing call info.

What was Genesys thinking, seriously...