Hardware LB are more performant than software LB.
Your vsd only shows a regular server group design, meaning a user connects to LB in front of mid tiers, the LB balances to one of the mid-tier behind it (one or the other, both are active), the query then goes to the second LB that balances the query to one of the 2 ARS servers behind him (one or the other).
In this vsd, all servers are active and working, no need to chqnge IPs and they're connected on the same database.
Let's call mid tiers servers MT1 and MT2, ars servers ARS1 to ARS3 (the one not accessible from LB).
ARS3 is in charge of administrative tasks, heavy operations (reconciliations, fts indexing, emails...).
If MT2 is down, LB in front of mid tiers will see that (if you coded some life control) and only route queries to MT1, hence no outage, only a possible slowr response time since MT1 would have to handle more connections.
If ARS1 is down, LB in front of ARS servers will detect it (if you coded some life control) and will only forward queries to ARS2, same principle.
If ARS3 is down, if you defined ranking form properly, the operations he handled would be handled by the next ars server defined in the ranking form.
In your case I guess you "just" have a mid tier and an ARS right? the other servers are backups running on another database, possibly backuped up live?
Thanks Laurent for your very good explanation, according with our previous understandings, that is the state that we must pursuit. But Let me clarify some points:
Firstly, I upload a.vsd file with a generic recommended architecture for a small deployment.
Currently, I upload a new file (ARS.pdf) with a closer actual environment diagram.
The initial provider builds this implementation without the load balancers (due oblivion (forgetfulness) at procurement / purchase process). So, currently, on this active-passive schema the failover processes must be made manually, at AR System and Mid-Tier levels.
We are looking for changing forward in order to provide active-active functionalities, at both levels.
So we need help in order with a manufacturer / model /reference /part number recommended and tested as hardware Load Balancer?
Also I want to know if AR Server Group configuration can provide HA functions as Automatic Failover Recovery and/or Load Balancing. Without HW load balancers, neither using OS clusters (Sun/Solaris in our case)
Additionally, we want to hear about any experience (supported by BMC) about software configurations for HA (Automatic Failover Recovery and/or Load Balancing) for Apache/Tomcat Services. Without HW load balancers, neither using OS clusters (MS-Microsoft in our case).
Well you could contact BMC for those kind of "real life" hardware, my last project involved Big IP F5, worked ok.
Something you could do in the mean time is something like this, you are in a server group, but since you don't have a LB in front of ARS, one Mid-tier is linked to an ARS "directly".
Instead of having a LB before the Mid-Tier, it could be an apache reverse proxy / loadbalancer, I saw some customers use this, but don't have a clue of configurations.
If MT 1 is down, LB "Apache" should be able to ping it and balance to MT2.
If ARS1 is down of course it's a problem since MT1 is connecter to ARS1. Easy way would be in MT1 to change the "hosts" file and change the IP / DNS so it points to ARS2.
Not a perfect solution but it would be better already.
I would want to go with the following appraoch.
1. Have a LB (Big IP F5) in front of the mid tiers
2. And have another Virtual LB between AR and Mid Tier connection.
This is a tested methodology by BMC for more their customers. They have even implemented this in their OnDemand Solution.
There are certain configurations which needs to be made in the host files of mid tier servers and ar.cfg file. Putting servers in HA also involves few API level configurations.
Let me know in case you want to go for this approach and I will share the details of the configurations with you.
That sounds good. So please let us know the details about that configuration.
We are reviewing “Hardware LB”:
And also interesting “Software LB” solutions like:
But, at the end our big concern continues being the official position of BMC about “supported” configurations.