|
Good day,
I'm new to virtual server architectures and am curious
if LVS could be put in place of my current adhoc
architecture. I figured I'd start with the experts before
I go off to implement a random architecture. Any way
here's quick run down I what I currently have.
I have a CPU intensive analytics web application which
is currently handled as follows:
One beefy Linux box with redundant power, RAID, etc acting
as the single point of access to the public network. This is
more than sufficient to handle our current hosting
needs. This box uses java servlets as a application platform.
This thing is up all the time, unless there is a severe silicon meltdown.
Connected to that box I have couple NT calculation boxes were
the CPU intensive chores are handled. Load balancing is handled
in the following maner: The linux and NT boxes communicate with
a simple tcp/ip protocol. The NT servers are iterative and stateless.
At startup the linux box establishes one connection to each of the
NT servers and MAINTAINS that connection unless there is a
system failure. The linux box has a single request queue. As
users request calculations they are placed in the queue. The
request are pulled from the queue and acted on by a thread
which handles communication with one of the NT servers.
Generally this works pretty well, but I'd really like to push the
load balancing chores off to a standard solution such as LVS,
but don't quite know if it is right for me. I believe the NAT solution
would be my only option since I currently don't speak a well known
protocol from the NT boxes, but it doesn't seem like the load balance
effect could work unless a connection was reestablished for each
request from the Linux box for each request. Is this true? This seems
pretty costly, but don't understand how this would work otherwise.
Could I use LVS without changing my architecture drastically?
TIA,
~christopher
|
