|
Hi Guys,
I apologise in advance if this seems obvious to you, but this is my first
attempt at installing/configuring LVS.
I understand most of the How-To so far ! well...the theory anyway!
The confusing bit is 'where to start' really.
Here's how my home network is conf'ed:
Cable modem --> Debian etch box setup as a gateway (2NICs: eth0 dhcp, eth1 =
10.1.1.1) --> switch --> workstation 1 (.200)
==> Workstation2 (.201)
==> Workstation 3 (.202)
==> ...etc
==> Server1 (.205)
==> Server2 (.206)
==> Server3 (.207)
==> Server4 (.208)
Now i understand that this is more than enough to test LVS.
What i intend to do is use Server1 as the director and server2,3,4 as the
realservers.
In terms of forwarding type i assume LVS-NAT is probably the easiest/more
convenient with regards to my systems.
I will install a bare install of Debian etch on all 4 servers (1dir +
3realservers).
As i understand it, the latest kernel used by Debian supports ipvs.
What i want to achieve at first is a very simple highly available cluster
serving one website (HTTP).
Ie: browse to a website, see which node i'm connected to, unplug the node
and see if i can still see the website.
I hope this is enough info for you to picture what i have and where i want
to go :-)
Here are a few questions to put me in the right direction if you don't mind
(it's more logistical than technical stuff really...)
Can you confirm that i should follow these steps in order:
1- Build the machines with completely bare OS installs
2- Configure the director
3- Configure each realserver
One thing i'm not sure of is: if the 2.6 kernel is already patched to
support ipvs, do i simply need to install ipvsadm (i checked and there's a
ready made deb package for it) to manage ipvs? No silly kernel recompile to
do?
Also, since i'm gonna go for a LVS-NAT forwarding method i understand i
won't have to deal with the arp issue. Is that right?
In the above setup, can i use my debian gateway server as the client to test
from ? I'm not too familiar with routing stuff or iptables rules and am a
bit worried that my lack of knowledge in this domain will screw my tests if
i try to test from an outside client.
Now a quick couple of questions about the configure script and other tools
provided in the howto:
I want to start with one director only (for simplicity...since i'm a noob),
but my ultimate challenge will be to implement this at work with a fully
resilient solution (2 directors failover). Is it easy to just add a director
to the equation in the future? Do you recommend UltraMonkey? Or am i going
to over complicate things with that?
I'm a bit worried that the quite is really old and that some new config
scripts (that would make my experience much easier) are out there and i
don't know about them :-). Any advice/scripts would be much appreciated!
Anyway,
Thanks a gazillion for your precious time.
Looking forward to reading you on the list.
Raph
|
