[cgl_discussion] Re: [Clusters_sig] A couple of questions

John Cherry cherry at osdl.org
Wed Jan 19 09:12:16 PST 2005


On Wed, 2005-01-19 at 13:53 +0100, Lars Marowsky-Bree wrote:
> On 2005-01-17T16:15:42, John Cherry <cherry at osdl.org> wrote:
> 
> > Cluster-aware Volume Management for Shared Storage
> > 	evms?
> 
> Yes, both Linux HA and the (alas, proprietary) OES Novell Clustering
> Services do support EVMS in cluster mode. The later one for up to 32
> nodes, I'm told.

I was only addressing the open source solutions and I have only seen a
data sheet for the OES Novell Cluster Services.  They claim "integration
between the cluster volume broker and EVMS".

BTW, the data sheet for Linux HA says that Linux-HA has no special
shared disk requirements.  There is no mention of EVMS anywhere on the
project site.

> 
> > Cluster File System
> > 	OpenAIS - no
> > 	Linux-ha - no
> 
> Wrong way to look at this I think, Linux HA is user-space only and thus
> of course doesn't include a CFS, same as OpenAIS. They ain't aiming at
> providing a full stack on their own.
> 
> Using OCFSv2 with Linux HA would address this, for example.

Agreed.  Things like cluster filesystem, cluster volume management, and
network storage replication may not be part of a core clustering
solution...especially a portable, user-level clustering solution.  I
probably should have just listed open source cluster filesystems such as
OpenGFS, OCFS, Lustre, etc.

However, services such as a distributed lock manager and/or cluster
messaging could be considered to be common cluster services that would
enable the management and coordination of shared resources.  Wouldn't it
be great if these services were just part of a standard Linux
kernel?  :)
 
> 
> > Cluster-wide Log Collection
> > 	???
> 
> syslog(-ng), evlog...

These are building blocks for cluster-wide log collection, but I would
not call them solutions for them yet.

> 
> > SW Rolling Upgrades
> > 	OpenAIS - ??
> > 	Linux-ha - ??
> 
> Rolling upgrades and updates are supported with Linux HA for the cluster
> software itself.

Yes.  This seems to be the definition of rolling upgrades nowadays.
True SW Rolling Upgrades is a system software stack issue...not to
mention data consistency issues.

-- 
John Cherry
Open Source Development Lab
CGL - Roadmap Coordinator




More information about the cgl_discussion mailing list