Simple Network Filesystem with redundancy / failback feature

English Network/Internet
#1

Dear Group,

I supervise a couple of machines and still look for a
better all round solution concerning user management and network shares.

Ideally it should be:

  • simple to (re)setup
  • secure in sense of having HA features like supporting redundant servers
  • there’s no downtime when one machine crashes
  • a server can be updated/maintained without downtimes of the whole network
  • stable for production use
  • good performance / throughput
  • ideally natively supported by the kernel, no haze with compiling, fixing, version conflicts
  • secure in sense of that it should be difficult to break in
  • Free of charge (no enterprise edition needed, no proprietary solution)
  • Ideally users can be taken over from old passwd / NIS

I am using SuSE for quite a while but have not found appropriate solution.
Has anyone good experience and a 100% bullet proof solution for the use case?

I tried:

  • NFS3 and NIS that’s the way it works now, easy to setup, fast, well supported, insecure, no native redundancy
  • AFS / Kerberos + LDAP, difficult to setup, lot to learn before starting, secure, redundancy support (but did not get it to work), lot’s to adapt in SuSE to make it work, inconsistent documentation, Debian is much more comfortable with regard to it
  • NFS4 + Kerberos + LDAP, difficult to setup, fast, inconsistent documentation, no redundancy, Debian is much more comfortable with regard to it
  • glusterfs, has all features I need, but immature, slow if modifed fuse module is not used
    (conflicts with the one of SuSE), buggy
  • Samba, slow, Windows is also supported, did not spend much time on it

Kind regards,

Josh

#2

On Mon, 2009-09-28 at 07:26 +0000, josh suser wrote:
> Dear Group,
>
> I supervise a couple of machines and still look for a
> better all round solution concerning user management and network
> shares.
>
> Ideally it should be:
> - simple to (re)setup
> - secure in sense of having HA features like supporting redundant
> servers
> - there’s no downtime when one machine crashes

That’s a cluster. Really no other way to satisfy that one.

> - a server can be updated/maintained without downtimes of the whole
> network
> - stable for production use
> - good performance / throughput
> - ideally natively supported by the kernel, no haze with compiling,
> fixing, version conflicts
> - secure in sense of that it should be difficult to break in
> - Free of charge (no enterprise edition needed, no proprietary
> solution)
> - Ideally users can be taken over from old passwd / NIS
>
> I am using SuSE for quite a while but have not found appropriate
> solution.
> Has anyone good experience and a 100% bullet proof solution for the use
> case?

NOTHING is bullet proof. Whoever told you that was lying to you. I do
think a cluster is the closest to what you are looking for. Why? You
can actually have a shared filesystem (e.g. ocfs2). NFS isn’t
sufficient because it doesn’t provide cache/write consistency across
nodes.

>
> I tried:
> - NFS3 and NIS that’s the way it works now, easy to setup, fast, well
> supported, insecure, no native redundancy
> - AFS / Kerberos + LDAP, difficult to setup, lot to learn before
> starting, secure, redundancy support (but did not get it to work), lot’s
> to adapt in SuSE to make it work, inconsistent documentation, Debian is
> much more comfortable with regard to it
> - NFS4 + Kerberos + LDAP, difficult to setup, fast, inconsistent
> documentation, no redundancy, Debian is much more comfortable with
> regard to it
> - glusterfs, has all features I need, but immature, slow if modifed
> fuse module is not used
> (conflicts with the one of SuSE), buggy
> - Samba, slow, Windows is also supported, did not spend much time on it

Doing a cluster the SLES way (openSUSE as a enterprise level HA??? I’m
not sure I’d recommend that) would be to use ocfs2 as the cluster
filesystem with HA heartbeat.

I have set these up in the past. Are they reliable? They’re ok… I
think more work is needed frankly.

I also use Veritas, which has limited platform support, can’t just run
on any ole Linux. It’s VERY expensive though.

Red Hat also has documentation on their clustering solution (using
RHEL). It might be something to look into as well.

#3

Hi,

thanx for inspiration.
I’ll take a deeper look into heartbeat / drbd
maybe that could be the building base to something
useful.

SLES or RHEL is no answer to the problem for me
cause it doesn’t simplify the whole case proportionate
to the cost. And I don’t want to rely on proprietary stuff
of novell. If I wanted that I could also very well switch to
a monolithic windows environment.
I also tried CENTOS (= Free flavor RH) but I want to keep the amount of distributions manageable.
If all else fails I’ll try this one.

With Bullet Proof I mean - I follow the instructions and it works. No stuck at a point. For server concerns Debian seems here very ideal. But most of the clients are OpenSuse.

This is no unusual use case. All I want is redundancy for my infrastructure.
I have UPS, will have soon a good SAN for the raw storage, all I need is a robust redundant file server and usermanagement solution.

Bye,

Josh