Help !! Clustered Servers

Hallo le Cat,

I think that sounds like a lot of dough for that kind of metal, we had to install a simliar solution in our office and I don't belive it cost so much. I could perhaps ask our specialist for details. Regarding servers, there is a lotof info available at the CeBIT, and there you could certainly find some experts, also electronically. Try HP's site on the fair:
http://www.hewlett-packard.de/cebit/index_eng.html
I'd look there if I were you. Good luck!
Zara

Why should an organization need Clustered Servers?
Maybe for supporting a new development!
Maybe for using some extra floor space and extra power in the new UPS!
Maybe ...., maybe ...., maybe ....
If you ask the executive level in the company they should tell you something like : "clust-what?; all I want is the XYZ project running on time, with good performance, and without worring about system failures".
So, if you need to set-up a cluster you better use good hardware, choose the right platform, and give Scalability, High-Availability and Performance it's place in the project goals!
U$D 30000 for a cluster? is too much for a cheap solution (maybe you could have similar MTBF with good hardware and software without going to clustering). But U$D 30000 is nothig in professional systems clustering.
You should first check your needs very carefuly.
Look at IBM (www.ibm.com) for info abou clustering, it may be usefull.

Regards,

Jose P. Mir.-
jpm@jpmir.net

Clustered systems are expensive. $30k does NOT sound like a lot. I would take a good look at the solution and make sure it's not too cheap. Also, 36GB seems awfully small.

The concept of a cluster is simple: 2 servers sharing a single, external RAID enclosure. They also have a dedicated link between the two servers so they can detect if their partner fails. In a Microsoft SQL Cluster, you generally split the RAID between the two servers. Lets say it a 200GB RAID. You create 2 logical drives: G: 100GB and H: 100GB. You assign G: to ServerA and H: to ServerB. Under normal conditions, ServerA will see only the G: drive and ServerB will see only the H: drive. You install 2 instances of SQL on EACH server and name them, for example SQLA and SQLB. SQLA expects to find its data on G: and SQLB expects to find its data on H:. On ServerA, the instance of SQLA will be running and the instance of SQLB will be stopped (these will appear as seperate services). On ServerB, the instance of SQLB will be running and the instance of SQLA will be stopped. Let's assume ServerB fails. ServerA notices immediately that ServerB is not responding. He sends a signal to the RAID enclosure and takes over the H: drive. Now, ServerA has both G: and H:. The cluster software will kick off a process (like a batch file) that will start the SQLB instance running on ServerA which finds its data on H:. The client application is using the SQL instance name to find the data so after a short time-out, the client will reconnect to SQLB not realizing that it is now running on a different server.

As for how reliable they are, that depends on the hardware (don't go cheap) and how well it's configured. Remember, the hardware of EACH server needsto be able to handle the combined load of BOTH servers. Otherwise the server will chock on the load when his partner fails. Under the right circumstances, the results are awesome.

It seems to me that the main question is not about how to partition the disks or launch the processes. Before planning that is necessary to understand the different options and evaluate the needs. Suggesting the size of the hard disks is, at this stage, a wrong indication about the way to follow. Maybe is very clear for experienced people that disk space is never too big, but, this is not relevant to the clustering implementation.
I think the purpose of this Q&A is to give guidelines, not to “teach”. I know the intentions are the best in all of us, but we must keep in mind the fact that our experience and/or knowledge may not be 100% exact, or even if it is, maybe we are not the better communicators. For example, is not true that “Clusters are 2 servers sharing a single, external RAID enclosure”. Clustering is much more than that; and even if I know that this answer could be right in many situations, I must disagree with it because could give the wrong picturein many other situations.
IT is not a self-containing human activity. IT is deeply immerse in a business environment. The solutions that IT implement are implemented to solution some organization’s needs; so it’s necessary analyze theseneeds to find the right solution. With the info we have in this question nobody can give a serious answer; we only can give good information to let “leehellcat” find out the right solution.

Jose P. Mir
jpm@jpmir.net

Clustering is a technology that looks to improve Availability and Performance. Different platforms and implementation ways exist to deploy clustering solutions.
Microsoft defines two flavors of clustering. The mere Clustering and the Network Load Balance. You could find useful information about these topics in Microsoft’s web site, or, if you want, I can mail you some documents (downloaded from MS).
Red Hat, and other Linux manufacturers, also have clustering solutions. In the Linux world you could have vantages and disadvantages you need to evaluate carefully.
IBM has good clustering options in his proprietary AIX OS; and if you select the deployment of the system in the new pSeries servers you will see a extremely strong infrastructure that gives, in the long run, peace of mind.
Oracle, with it’s new 9i version, is pointing to (Real) Application Clusters, and gives good comparisons against IBM and Microsoft clustering solutions. Of course, you could deploy Oracle9i over Win2k or AIX.
Don’t discard the outsourcing of the site hosting, at least at the beginning while you are evaluating the real needs and check technology solutions it could be useful the flexibility and external support.

Jose P. Mirjpm@jpmir.net