Cluster questions

First of all, excuse me if my English isn't too correct.
I need to install a portal infrastructure that was developed with ASP
and SQL Server. The system would like to be redundant and scalable.
To do that, I have in mind, the following diagram :
- 2 W2003 Servers with IIS6 and NLB ( frontend )
- 2 W2003 Servers with MS-SQL and MSCS with a SAN connected ( backend
).
I have read Microsoft and others documentation, and I have questions
that I would like to know.
I have worked with NLB and I understand how it works, but I'm new with
MSCS, and I must need to know to cluster SQL servers.
In SQL 2000, the cluster can be active/pasive or active/active.
If it is active/pasive, one of the two nodes, assumes all work, and
the second node is waiting that the first node goes down, when that
happens the second node assumes all work, this is the failover. Every
node has got one IP and the two nodes have a IP that they share (
Virtual Server), like NLB.
The ASP applications connect to the IP of Virtual Server. No problem.
The cluster is transparent to the ASP aplication.
But if I have active/active mode, the application must see two SQL
Server, not one like active/pasive mode, it's true ? The cluster isn't
transparent to the ASP aplication ?
The developement of the ASP application is different that the
active/pasive and active/active modes ?
With MSCS is possible to make the system scalable or only the failover
is possible ? If it's possible the scalability, is necessary to modify
the ASP developement or it's transparent ?
I read that MSCS is necesary that the servers must be joined in a
domain, it can't work in a Workgroup.
If I install AD in the 2 web servers, the performance will decrease,
there won't be users and permissions, it's only a web portal.
Is it possible to work in a Workgroup ?
And at the end, give me your opinion about :
- Install SQL 2000 SP4 or SQL 2005 ?
- Install MSCS or a third party software, like Computer associates ?
Thanks in advance,
Xavi.
You seem to have a misunderstanding of what MSCS does and how it works with
SQL. That is not unexpected since the terminoloty is very confusing, even
to clustering veterans.
Here is an excellent starting point for understanding SQL Clustering.
http://www.microsoft.com/technet/pro.../failclus.mspx
The document describes SQL 2000 clustering but it applies to SQL 2005 as
well.
As for your questions, Active-Active and Active-Passive are no longer valid
terms to describe SQL Clusters. You now have single- or multi- instance
clusters. It sounds like you are looking for a single instance cluster with
two nodes. Aclustered instance looks just like any other SQL instance from
the client. There is no difference from an application perspective whether
a SQL instnace is clustered or not. An application should handle brief
database connection disruptions gracefully in case there is a cluster
failover, but that is part of good application design anyway.
If I was starting from scratch, I would use SQL 2005.
MSCS does require a domain and will not work in a Workgroup. Anytime I
build a highly available application system, I prefer to build a monitoring
and management console as part of the system. That can be your primary AD
controller. I also use it to manage backups and system monitoring/alerting
tools.
Geoff N. Hiten
Senior Database Administrator
Microsoft SQL Server MVP
"xavi" <xavi@.comb.es> wrote in message
news:1130854797.337258.155860@.g49g2000cwa.googlegr oups.com...
> First of all, excuse me if my English isn't too correct.
> I need to install a portal infrastructure that was developed with ASP
> and SQL Server. The system would like to be redundant and scalable.
> To do that, I have in mind, the following diagram :
> - 2 W2003 Servers with IIS6 and NLB ( frontend )
> - 2 W2003 Servers with MS-SQL and MSCS with a SAN connected ( backend
> ).
> I have read Microsoft and others documentation, and I have questions
> that I would like to know.
> I have worked with NLB and I understand how it works, but I'm new with
> MSCS, and I must need to know to cluster SQL servers.
> In SQL 2000, the cluster can be active/pasive or active/active.
> If it is active/pasive, one of the two nodes, assumes all work, and
> the second node is waiting that the first node goes down, when that
> happens the second node assumes all work, this is the failover. Every
> node has got one IP and the two nodes have a IP that they share (
> Virtual Server), like NLB.
> The ASP applications connect to the IP of Virtual Server. No problem.
> The cluster is transparent to the ASP aplication.
> But if I have active/active mode, the application must see two SQL
> Server, not one like active/pasive mode, it's true ? The cluster isn't
> transparent to the ASP aplication ?
> The developement of the ASP application is different that the
> active/pasive and active/active modes ?
> With MSCS is possible to make the system scalable or only the failover
> is possible ? If it's possible the scalability, is necessary to modify
> the ASP developement or it's transparent ?
> I read that MSCS is necesary that the servers must be joined in a
> domain, it can't work in a Workgroup.
> If I install AD in the 2 web servers, the performance will decrease,
> there won't be users and permissions, it's only a web portal.
> Is it possible to work in a Workgroup ?
> And at the end, give me your opinion about :
> - Install SQL 2000 SP4 or SQL 2005 ?
> - Install MSCS or a third party software, like Computer associates ?
> Thanks in advance,
> Xavi.
>
|||As you can see, I'm not a cluster veteran :-)
Thanks for the info, and the link documentation is very good.
I understand the single-instance failover cluster, but I don't
understand too much multi-instance cluster.
Is there no difference from the application prespective in a
multi-instance cluster ?
It is possible to do in the scenario descrived ?
Is there any example of a multi-instance cluster ?
Thanks,
Xavi.
|||Each extra instance looks like another SQL Server as far as the clients are
concerned. Each clustered instance has independent disk resources, IP
address(es), and unique network name. You connect to a Virtual SQL Server,
not to the underlying host node.
You need to differentiate between Nodes (host computers) and instances
(Virtual SQL Servers) in a cluster. They exist independently.in a cluster.
You can have up to 16 instances and eight nodes in a cluster (SQL 2005).
You get four nodes under SQL 2000. If you need multiple clustered SQL
Servers, you can create multiple instances on a cluster and save some money.
If you need only one SQL Server, then you create a single-instance cluster.
SQL Clustering is a failover technology, not a scale-up technology. You
cannot have two host nodes connected to the same database for scale out.
That is why the Active-Active terminology is misleading.
Your general idea looks good.
GNH
"xavi" <xavi@.comb.es> wrote in message
news:1130878710.578655.77290@.f14g2000cwb.googlegro ups.com...
> As you can see, I'm not a cluster veteran :-)
> Thanks for the info, and the link documentation is very good.
> I understand the single-instance failover cluster, but I don't
> understand too much multi-instance cluster.
> Is there no difference from the application prespective in a
> multi-instance cluster ?
> It is possible to do in the scenario descrived ?
> Is there any example of a multi-instance cluster ?
> Thanks,
> Xavi.
>
|||You can run SQL Server in NLB, which is known as a Federation of Servers.
Server clusters are for redundancy and high availability, not scalability.
If you are looking for scalability, which does provide some high
availability, you should look at server federations and distributed
partitioned views. This is most often used for partitioned VLB
installations, mostly warehouse, read-only type systems, rarely OLTP
systems.
Sincerely,
Anthony Thomas

"Geoff N. Hiten" <SRDBA@.Careerbuilder.com> wrote in message
news:OuR$Tw13FHA.3600@.TK2MSFTNGP12.phx.gbl...
> Each extra instance looks like another SQL Server as far as the clients
are
> concerned. Each clustered instance has independent disk resources, IP
> address(es), and unique network name. You connect to a Virtual SQL
Server,
> not to the underlying host node.
> You need to differentiate between Nodes (host computers) and instances
> (Virtual SQL Servers) in a cluster. They exist independently.in a
cluster.
> You can have up to 16 instances and eight nodes in a cluster (SQL 2005).
> You get four nodes under SQL 2000. If you need multiple clustered SQL
> Servers, you can create multiple instances on a cluster and save some
money.
> If you need only one SQL Server, then you create a single-instance
cluster.
> SQL Clustering is a failover technology, not a scale-up technology. You
> cannot have two host nodes connected to the same database for scale out.
> That is why the Active-Active terminology is misleading.
> Your general idea looks good.
> GNH
> "xavi" <xavi@.comb.es> wrote in message
> news:1130878710.578655.77290@.f14g2000cwb.googlegro ups.com...
>
|||I see that, in the main schema, the SQL servers that use MSCS must be
joined in a domain.
The Geoff N. Hiten suggestion about using an other server as a DC seems
a good idea, but what does it happens if this DC server fails? Does the
SQL cluster continue alive ?
What are the consecuences of a DC failure ?
I would not like using 2 aditional servers for 2 DCs just for DC
redundance.
Thanks,
Xavi.
|||Hi
You loose your DC, your cluster goes down.
If you need the resilience of a cluster, and can afford all that expensive
hardware, having small servers as 2 DC's is not much to ask for.
Regards
Mike Epprecht, Microsoft SQL Server MVP
Zurich, Switzerland
IM: mike@.epprecht.net
MVP Program: http://www.microsoft.com/mvp
Blog: http://www.msmvps.com/epprecht/
"xavi" <xavi@.comb.es> wrote in message
news:1133345191.463210.32350@.g47g2000cwa.googlegro ups.com...
>I see that, in the main schema, the SQL servers that use MSCS must be
> joined in a domain.
> The Geoff N. Hiten suggestion about using an other server as a DC seems
> a good idea, but what does it happens if this DC server fails? Does the
> SQL cluster continue alive ?
> What are the consecuences of a DC failure ?
> I would not like using 2 aditional servers for 2 DCs just for DC
> redundance.
> Thanks,
> Xavi.
>