On Tuesday 10 February 2009, Zahid Mehmood wrote:

I'm wondering about how people are managing the subscriber and

other application data in a large environment? What are people using to share data among the servers behind the dispatcher? Shared database? db replication?

Hi Zahid,

some people use a shared DB, e.g. mysql cluster. Some also implement some database partioning on top of normal mysql in their server. DB replication is also frequently used, to place your read/only data (auth, groups..) on local host to minimize network overhead.

A central location for collecting accounting records? ( I know people don't usually rely on proxy's accounting records but i'd still like to collect that information)

Accounting data can be easily aggregated over many servers in one DB, as each call is normally unique for the whole system.

Also, What is the best way to update the location information on server that was offline for sometime?

If one of your DBs was offline for some time, then you need to wait until the registration is expired for read access, but continue to write to it. When the registration expire time is over, then you can start to read again.

Cheers,

Henning

Hello,

On 02/10/2009 07:55 PM, Zahid Mehmood wrote:

Thanks for your reply.

May I trouble you and others on the list to share the deciding factors (pros & cons) for choosing the option that you did.

the benefit of a cluster is that it was designed for such cases. The drawback is that MySQL version still is pretty early stage. However, there are success stories and production environments with the cluster in place.

I am particularly going for user partitioning. It does not make full service unavailable if one db node is down. When partitioning is very important to group users by probability to talk inside the group.

As example, put all IT guys in the same node, they are very likely to talk in between a lot to solve issues. Of course, such distribution can be considered on geographic location if you deal with branch offices, a.s.o.

Each node is replicated for HA and failover - active-active or active-standby, depending on needs.

The communication between nodes is another part. Depending on number of nodes, you can have static binding of users and nodes or a proxy node that just gives back the node of one user. So you look for user on local node and if not found ask the proxy node.

I am designing and implementing the platform mainly based on customer expertise. If they feel more comfortable with a cluster, that is. Practically they are the ones that have to manage the system in a day by day manner and integrate with third party components.

I'm trying to get a better understanding circumstances in which you may choose one over the other.

What, if any, special considerations are there for the presence part?

Presence has more sip traffic, requires more memory to store and aggregate the presence documents. Therefore I recommend you have dedicated nodes as presence servers. It helps a lot in scaling and ensuring service availability for basic telephony (registration and call routing).

Cheers, Daniel