Topic: Cluster e-smith installations?

[Kevin Baker]

Has anyone tried clustering a group of e-smith machines?

I am very interested in growing the base e-smith install into a fully clusterable network server similiar to sun cobalt server appliance.

[ed sharpe]

sounds like an interesting project!
hook 'em together with fibre!
Alas, I do posess the talent to do it...
but I am sure someone out there does!
ed!

[Johan]

Hello Kevin and Ed,

On the page's of RequestedDeletion is discribe the follow project for SME.

Linux High Availability How-To for Mitel SME v5.1.2

De link is http://www.star-support.com/sme/Linux-HA/SME%20High%20Availability%20How-To.html

I think is that where you searching for two are more server whith fall over.

Grtz

Johan

I have made in the earlier days a test case for this and it realy works how its discript. But we don't need't for are company.

[ralph]

Hi Everybody,

what RequestedDeletion describes is the currently most basic way of clustering linux machines using drdb.
It's cool, it's simple and it works - even over longer distances.

What - in my mind - it is NOT intended to be used for are so called active/active clusters. If you are looking for something like that, go to www.suse.com and search for failsafe. Pro : active/active  CON ( for some ) higher cost due to the need of shared storage ( scsi or fibre ).

Finally : Still looking for the ultimate clustering solution for sme ( for quite a while now ).

regards,

Ralph

[guestHH]

Hi Ralph,

Could you explain a little more about what you want to achieve ?
There are _many_ projects on the net for linux clustering, what I would like to know what _your_ definition of 'clustering' is.


Regards,
guestHH

[Kevin Baker]

My definition of clustering....  well nothing crazy, to be very simplistic: a bunch of machines working together over a network, functioning as one single machine.  These machines should collectively provide load balancing for performance boosts, scalability to grow with an application's system requirements and high availabity for a more reliable server solutions.

You are right there are a number of possible solutions out there.  My main goal would be to provide a simple blade that installs one of these solutions on e-smith.

I am currently working on a simple procedure for installing JBoss 3.0 on an e-smith machine.  My hope is that this in combination with e-smith will be a solid platform for our J2EE compliant products, that are reliant on both JBoss and the other services already provided by e-smith (http, ldap, imap, smtp).  I have a ways to go though.  JBoss provides clustering services for its J2EE server solution.  Great stuff.


JBoss clustering article.
http://www.onjava.com/lpt/a//onjava/2002/07/10/jboss.html

[Kevin Baker]

My definition of clustering....  well nothing crazy, to be very simplistic: a bunch of machines working together over a network, functioning as one single machine.  These machines should collectively provide load balancing for performance boosts, scalability to grow with an application's system requirements and high availabity for a more reliable server solutions.

You are right there are a number of possible solutions out there.  My main goal would be to provide a simple blade that installs one of these solutions on e-smith.

I am currently working on a simple procedure for installing JBoss 3.0 on an e-smith machine.  My hope is that this in combination with e-smith will be a solid platform for our J2EE compliant products, that are reliant on both JBoss and the other services already provided by e-smith (http, ldap, imap, smtp).  I have a ways to go though.  JBoss provides clustering services for its J2EE server solution.  Great stuff.


JBoss clustering article.
http://www.onjava.com/lpt/a//onjava/2002/07/10/jboss.html

[guestHH]

Hmmm, it takes more then a 'simple blade' to achieve what you want.

Clustering in this case means:
- getting a robust journaling file system working with the currunt kernel
- hook up to an external redundant storage
- balance services which would require a kernel re-compile
- balance workload with e.g. LVS, unless that Jbos thing has balancing capabilities within itself.

I think too many adjustments to the base SME server will be required, to get this running.

Regards,
guestHH

[ralph]

Hi RequestedDeletion,

sorry, took a while for me to get back to this ...

Here's what I'd like to achieve :

- have a group ( at least 2 nodes ) of sme servers running in an active/active cluster
- use shared storage ( pref. fibre )
- use cluster for all services ( mail, file sharing, printing, apaching, ... )

I do know that it sounds much more simple than it actually is. Having had a look on several clustering projects out there in the community, I havent yet found a proper one.

Despite that I'm still fighting with the installation of sme 512 on a external array connectet by fibre channel using either qlogic 2100 or emulex lp8000 fc hba.

Getting the fibre stuff to work might be a step towards using it as shared storage for maybe hearbeat or failsafe based cluster.
Even doing so on two identical boxes using drdb would be a step ahead - by the way : you did an excellent job on your how-to !!!!

So far so well, think the rest will be some more struggling around.

On the other hand : any ( !!!!!! ) help will be heavily appreciated !!!!

br, ciao,

Ralph

[guestHH]

Hi Ralph,

Depending on your (future) expected load on the external storage, I saw some really cheap (compared to scsi or fiber) IDE solutions. Don't remember the brand by heart (3ware???).

1. In order to monitor the independent services (mail, smb etc) you need to install MON (http://www.kernel.org/software/mon/).
I did not play with it yet, but you'll need it.

2. There can only be 1 active server, so load balancing can not be done. 'Just' Hot Stand-by for hardware and individual services*. (*in case of external storage only)

3. Maybe you can direct MON to switch 1 service from the primary server to the seconday server when the load on the primary get's too high (acording your settings). You have to study MON for that. (let us know )

Regards,
guestHH

[ralph]

Hi RequestedDeletion,

I'll have a time out for round about two weeks on that topic ( lots of work ) but as soon as I'm back with any results, I'll let you know.

br, ciao,

Ralph

[guestHH]

Hi Ralph,

looking forward to your experiences.

Regards,
guestHH

[ralph]

Hi RequestedDeletion,

got to ask you for some patience on our topic - I actually do not have the time to play around ( life is hard somethimes ).

Hold on anyway and - if there's any news from you, let me know ( just mail it to me ).

Have a good time,

cheerio,

Ralph

[guestHH]

Hi Ralph,

No problem. We all know how it can be...

Have fun.

Regards,
guestHH