Subject: | |
From: | |
Reply To: | |
Date: | Wed, 23 Apr 1997 07:37:00 -0500 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
In article <[log in to unmask]>, [log in to unmask] (aar
on t porter) writes:
> What the heck is clustering?
>
> I had thaugh that the idea was to provide semi-redundant services
> through co-operative machines. Based on recent Miavx1 problems, that
> doesn't seem to be the case. I'm just really confused.
Clustering allows several systems to share resources. If configured
properly, they allow multiple systems to offer the same services in a
redundant and load balancing manner.
This does not mean that an interactive login on one system will
automatically fail over to another system if the first one fails.
Clustering should allow the user to relogin on the redundant node, though.
So why doesn't this work at Miami? Primarily, because the world is now run
by TCP/IP which has no concepts of redundancy or load balancing for
services. A domain name only points to 1 system (ok...it can point to
multiple systems...but is only served in a round robin fashion...not load
balanced or allowing for a down system).
Miami's clustered systems do share resources though (single
username/password database, shared tape drive, shared disk drives, etc.)
We've specicialized our systems for various functions. Therefore, we can
come closer to guaranteeing response time for certain operations (such as
network services on STREAM).
--
Kent Covert, Software Coordinator
Miami Computing and Information Services
Miami University, Oxford, OH
[log in to unmask]
|
|
|