Load Balancing 2 web servers after removing the ACE

[dvbell]

We have 2 internal web servers that are being load balanced round robin by an ACE. The ACE is going away and will not be replaced with an appliance.

My boss would like to have the 2 internal web servers receive the http requests and decide between them which will actually handle the request in a round robin configuration without adding an additional server or appliance into the equation. This would eliminate a single point of failure, the appliance, and if one of the servers goes down the other will continue to service the http requests.

I'm told that Windows does NLB with a few clicks between 2 servers and doesn't require using a third device like the ACE.

Is this possible to configure on Linux? Is there software to make this happen or can it be done with only the OS? We are using SUSE SLES 11 SP2. Thanks.

[Irithori]

Hi and welcome.

Maybe your boss tries to cut costs and listened to only the information supporting that

For starters: NLB is the poor manīs loadbalancer.
It works by having the same IP on multiple nodes, so each nodes getīs all packets.
Then there is comunication and logic between them to determine, who will answer.
That approach is ok-ish, if you have CPU power to burn.

But the configuration possibilites are way behind what a full-featured LB such as
the Singray Taffic Manager (former Zeus) or The Linux Virtual Server Project - Linux Server Cluster for Load Balancing can offer.

Also, there is no monitoring of the balanced services with NLB.

Why does NLB work with as low as two machines?
Because it runs on the same nodes. (Obviously )
This can be technically also be done with LVS.

But there are strong reasons against this, mainly different incarnations of "separation of machine purpose"
- Load metrics: You cannot easily tell, if the LB or webserver or a background job is responsible for load peaks
- Complex config: Especially network config will be not simple and therefore errorprone
- Update cycles: With every update -be it on the webserver or Lb side- the whole construct is endangered
- Security: With all services "on the frontline", they can be more easily targeted.
- Availability: Every machine down affects LB and Webserver, hence extra care and maintenance is needed to restore it


So in short:
A few new machines with either LVS or STM
or appliances for the same purpose are the much more saner route.

[dvbell]

Thanks for the input. I appreciate it.