Cluster - load balancing
support for cluster and load ballancing with internal and LDAP users
ACSO support commented
Active/passive would be a good start. It would be good have an easy fall back in case the VM that Kerio sits on detonates.....
Perttu Aaltonen commented
Active/Active or Active/Passive with shared storage would certainly be nice.
I would be happy even with Active/Passive on local storage if the servers would function in lock-step so that every change would be mirrored on the spare server. Activating the backup server could be done with something like VRRP or CARP.
Huge mail stores could be a problem as the storage need would double, but for a more manageable store sizes this could work well. Even with large mail stores it would probably be cheaper than to build a SAN. Passive host could also be used for doing offsite backups without slowing down the main server.
I think there a three possible cluster situations:
1. Active/Active Proxy (for distributed domains): You have one Ip address with a loadblanacer and several "frontend" kerios behind them. The loadbalancer choose a kerio. The user logs into these frontend kerio and will be redirected to the "backend" kerio with his datastore. If one "frontend" crashes, the loadbalancer will detect it and send all new connections to the other servers. The frontend server do the virus/spam detection, but have only a small number of mails in the queues.
2. Active/Inactive Datastore: There is n+1 kerioservers with a shared datastore in a distributed domain setting. If one kerio fails. the spare server take over the externel ip and connect to the filestore. Easy with nfs or cluster filesystems. more complex with shared san volumes.
3. Dynamic Account movement: In a distributed domain, all servers report there load and in a nightly action account will be moved from one server to another to get the same load on all serves.
Cristiaan Brans commented
Load balancing in my view would only become possible if Kerio (perhaps optionally) steps away from the 'file based' idea and stuff the data in databases.
File locking is preventing you from the type of clustering / loadbalancing the requesters are probably looking for.
IT Manager commented
Kerio currently has the ability to setup a "backup server", but it's rather odd to setup IMHO. I'd rather see true clustering to allow for fail-over in real time (or close to real-time). We've suffered one e-mail crash so far, and it was devistating to say the least. If I had a hot fail-over, that crash would not have impacted our business at all.
Cristiaan Brans commented
active active failover situation and active active load balancing (multiple servers managing the same domain and users)
probably letting the index files loose and start using maildir like folder/file management would be a great start for that!
Eduard Šturdík commented
servers would communicate among themselves and exchange data on the current load, number of active users, network load and connections. new connections would be routed to the server is least busy, or who has more free system resources