Hi Blues,
I am been thinking about setting BQ cluster with a small farm of Front-End
BQ Servers.
Theorically it is possible based on current setup code base with
any modification
Schematic:
eg -
LoadBalancer (Active) --- LoadBalancer (Hot Standby)
<--- Foundry or Sea Monkey --->
Turn on SSL sticky
Weighted Spread
|
|
|
core switch
|
|
_________________________________
| | |
BQ server 1, BQ server 2, BQ server 3 (Front-Ends)
| | |
? | | | ?
Slave node --Core Switch--Core Switch----Core Switch-- Slave node
(Additional) | | | (Additional)
________________________________
1)Mount Front-Ends to
/home
/usr/sausalito/codb
/var/db
on to Network Attached Storage via NFS hard mount
2)Disable dbrecover on other servers,
let BQ server 1 monitor CODB integrity
Howto: Adding BQ server into cluster
1) Install BQ Server - must be same version, same distro (usual way)
2) Remove all vsites and Administrators from BQ GUI.
3) Check /usr/sausalito/codb/codb.oids must be pointing to the Same index.
4) Turn off BQ services in all servers
4) move a copy of /home, /usr/sausalito/codb & /var/db to NAS - NFS
5) mount /home, /usr/sausalito/codb, /var/db from BQ servers to NAS - NFS
6) Turn on BQ services in all servers
Comments:
- Tar a post installed BQ binary from a just installed setup.
Then use that tar copy to setup, update other BQ servers
- In cluster mode, RPM or pkg management is tedious.
Therefore a flat tar installation is prefered.
- Basically, only Loadbalancer and NAS tiers are sandwitched onto BQ
servers to form 3 tiers BQ solution without any binary modification.
*** Slave nodes are additional, do not implement!***
KIV - Improvement:
- 2 tier loadbalancing: L4 and openmosix for slave nodes
Change load balancing logic to fair, round robin
And use of openMosix to distribute "SMP" processes across BQ Servers or
slave node servers that not mounting to ANY mount point.
- Slave nodes are a suggestion
- By Adding slave nodes to cluster require Kernel modification on
Front-Ends and slave node servers. These slave nodes share out
resources to Front-End servers to lessen Front-End resources
contraints. Much similar to SMP processes on HT CPU.
Cheers
patrick