[Ilugc] How to customize a pc to cluster
- From: raju.rajsand@xxxxxxxxx (Rajagopal Swaminathan)
- Date: Mon Feb 22 18:41:12 2010
Greetings,
On Mon, Feb 22, 2010 at 6:03 PM, subhojit ojha <subhojit.ojha@xxxxxxxxx> wrote:
Dear Shrinivasan,
? ? ? ? ? ? ? ? ? ? ? ?Thanks for ur reply, I have gone through the above
link but not understood fully.
Actually, I just need that whenever there is
high load coming to my server, automatically the application
or OS has to be transfer to cluster, I think it may be exactly
Load-balancing cluster ( I have not understood HA cluster)
also if my server is getting crash then also the parent OS,
child OS and applications has to be transfer to the cluster
( i.e; cluster PC) without stopping any OS or application.
I just need, my server should run always without any intervene.
As I mention before I m going to use one or two PC to act as
cluster, so plz suggest me how can I achieve it...I have
decided to use Red hat cluster suit but I don't know how
it will work with my Debian based server.
First off please go through the doc of Redhat cluster suite. Remember
it is sold as RHEL Advanced Platform with considerably higher list
price that "Standard" and you will need to buy two subscriptions as a
HA cluster by definition consists of more than one node and a lot of
redudndancy in just about every element in the stack -- be it active
active or active passive.
Ok. having built and administered few HA RHCS, let me attempt to
explain with my very little knowledge.
Clusters are used for different purposes. High Availability (HA), High
performance computing (There is one in IITM - Check it out if you are
allowed)
From your requirement it seems that you need HA cluster.
To ensure that the service does not go down if one of the member node
dies. This can be active passive or active/active (lot of
pre-conditions, ifs-buts in this case).
Please think as service as an NFS server/apache server/database
server/terminal server (s/server/service/g)
Think of cluster as a phantom device (somewhat like Virtual IP)
wherein the actual work is delegated the to one of its members.
Who delegates and how it is another subject in itself aka load
balancing (Checkout Linux virtual server or LB devices)
Once a service is under control of cluster, none of the individual
nodes can start it using /etc/init.d/<service> start. In fact it
should never be.
Hope this intro is sufficient to whet your appetite.
If you have understood the above please revisit the earlier links by
one of the very erudite members of this list. then come back with
further questions.
I would suggest you use centos to experiment. No debian at this stage
please. It will only confuse you. You will find a lot more people who
have implemented multi-node Redhat clusters so the probability of
getting answers is higher
And please note that *any* clustering is complex if you dont
understand the underlying concepts. Administering and troubleshooting
the cluster is non-trivial and requires a lot of time and effort
investment. It is not simple as dong a partition/install packages. It
requires more.
HTH
Regards
Rajagopal
Other related posts:
