[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-cluster
Subject:    Re: Clusterwide pids
From:       Lars Marowsky-Bree <lmb () suse ! de>
Date:       2001-07-18 5:55:54
[Download RAW message or body]

On 2001-07-17T17:40:18,
   "Albert D. Cahalan" <acahalan@cs.uml.edu> said:

> > This is a *huge* issue for HA.  From an HA point of view, process
> > migration is a definite evil and not even neutral.  It makes the
> > job harder, and generally makes the systems less reliable
> > (presuming greater code complexity makes the code less reliable).
> 
> From an HP point of view, process migration is usually evil.
> You really want to have one single-threaded process per node.

OK. You are all clinging to the status quo, which is good, because it means
you value stability ;-)

However, process migration _does_ offer nice features for some cases (manual
migration without downtime before scheduled maintenance and even in some cases
where the HA cluster notices that a node is about to fail, automatic load
balancing for some types of HP clusters).

If process migration was implemented in a clean and stable way, I see it as
"worth the effort". It is not an answer to everything, of course.

(Most of the issues for HPC are related to the fact that the cluster
interconnect is usually slower than accessing local resources - however, this
may not always be the case if the resources are available from all nodes
(common shared storage, shared network), or the cross bar is "fast enough"
(faster than the speed the resource can deliver))

Sincerely,
    Lars Marowsky-Brée <lmb@suse.de>

-- 
Perfection is our goal, excellence will be tolerated. -- J. Yahl


Linux-cluster: generic cluster infrastructure for Linux
Archive:       http://mail.nl.linux.org/linux-cluster/

[prev in list] [next in list] [prev in thread] [next in thread]