[Linux-cluster] Configuration Between Real Servers and GFS Cluster

[Diamond Li]

let's imagine,  in this diagram, client1 and client2 are sending
numerous requests to server1, but in meantime, there is no request
from client3 and client4, that means server1 is overloaded,  server2
is starving. So that is why I say we can't balance workload between 3
servers.



On Mon, Dec 28, 2009 at 6:39 PM, Gordan Bobic <gordan at bobich.net> wrote:
> Diamond Li wrote:
>>
>> thanks for your reply. I am trying to understand what you mean. for
>> instance, I have 6 clients and 3 GFS servers, they are divided into 3
>> different groups
>>    client1            GFS server1
>>    client2
>>
>>    client3            GFS server2
>>    client4
>>
>>    client 5           GFS server3
>>    client6
>>
>> So my question is how to balance the workload between these 3 GFS
>> servers since they are hosting identical data?
>
> By the diagram above, you're already load balancing it. Have clients 1,2
> mount off server1's floating IP, clients 3,4 off server2, clients 5,6 off
> server3. If one of the servers fails it's IP fails over to one of the
> surviving servers, and you'll get 4 clients on one server and 2 on the
> other.
>
> Note, however, that if different clients are accessing files in the same
> directories, you'll get lock contention and locks will end up bouncing
> between the servers, which will seriously hurt performance.
>
> Gordan
>
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
>